CN119124191A - A robot motion planning method and device inspired by spatial relation memory - Google Patents
A robot motion planning method and device inspired by spatial relation memory Download PDFInfo
- Publication number
- CN119124191A CN119124191A CN202411193677.1A CN202411193677A CN119124191A CN 119124191 A CN119124191 A CN 119124191A CN 202411193677 A CN202411193677 A CN 202411193677A CN 119124191 A CN119124191 A CN 119124191A
- Authority
- CN
- China
- Prior art keywords
- random
- node
- position information
- robot
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/34—Route searching; Route guidance
- G01C21/3446—Details of route searching algorithms, e.g. Dijkstra, A*, arc-flags or using precalculated routes
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Automation & Control Theory (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Manipulator (AREA)
Abstract
The invention relates to the technical field of robot motion planning, in particular to a method and a device for planning robot motion inspired by spatial relationship memory. The method comprises the steps of constructing a first random geometric figure in a configuration space of a robot through a K-NN algorithm by using a selective sampling strategy according to motion initial position information, motion target position information and map information, inputting the first random geometric figure and obstacle position information into an intelligent edge selector for edge screening based on a preset confidence threshold value to obtain a second random geometric figure, calculating a heuristic value through a memory perception predictor according to the second random geometric figure to obtain a node heuristic value, and using a microminiatable geometric figure according to the node heuristic value and the second random geometric figureAnd the module searches the optimal path to obtain a second optimal path. The invention relates to a high-efficiency and accurate robot motion planning method based on human brain space relation memory heuristic.
Description
Technical Field
The invention relates to the technical field of robot motion planning, in particular to a method and a device for planning robot motion inspired by spatial relationship memory.
Background
The motion planning is a key technology in the robot system, and has remarkable application prospect in various fields such as traffic systems, medical systems and the like. The goal of motion planning is to find a collision-free path in the decision space that connects the starting point and the target point and ideally minimizes the path cost. However, the real world decision space typically contains a large number of continuous variables as well as a high-dimensional unstructured space, which greatly increases the complexity of motion planning. Therefore, realizing a fast and reliable motion planning method in a high-dimensional continuous space is important for improving the practical application thereof.
Classical motion planning solutions can be broadly divided into two categories, search-based methods and sampling-based methods. The planning problem is conceptualized as a graph search problem based on the search method. Sampling-based methods, including fast exploration of random trees and probabilistic roadmaps, build exploration trees connecting starting and target points by random sampling in a configuration space. However, classical approaches tend to be inefficient in dealing with high-dimensional planning problems. In a high-dimensional continuous environment, a learning-based planning method is a key means for improving motion planning performance. Existing planning methods still present the following challenges in facing a high-dimensional continuous environment.
The traditional planning method builds a random geometric figure based on a uniform sampling strategy, namely, the whole space is explored indiscriminately, and the structural information of the actual problem cannot be fully utilized. Unnecessary node exploration or improper edges and nodes exist possibly, so that the method has poor effect in dealing with new problems and cannot be effectively popularized. When faced with fast or close obstacles, the robot must be able to quickly and accurately make decisions to avoid the obstacle. Conventional planning methods tend to focus on node location information and process each task independently, failing to take full advantage of past experience and established data models.
The human brain has natural advantages in the decision and reasoning process, where spatial relationship memory is critical in this process. Studies show that the human brain can convert the decision-making problem into a graph structure corresponding to the sensory observation, and the relation memory integrates the sensory observation with the structural relation of the sensory observation, so that sensory reasoning is promoted. On the one hand, humans can make complex inferences from sparse observations and make rapid decisions by integrating previous experience with current environmental information, which is related to the hippocampal-entorhinal cortex system. On the other hand, spatial memory is critical to human survival, and humans use "cognitive maps" to structurally understand the world, a process closely related to the hippocampus. Accordingly, to address the two challenges described above, the present invention contemplates a brain-inspired neural network (Graph Neural Network, GNN) -based planner to improve planning efficiency and scalability, referred to as brain-like motion planning (Brainy Motion PlaNNing, brainyMP).
In recent years, a learning-based planning method is the most promising method for improving the planning performance of high-dimensional continuous motion. By utilizing the strong learning and representing capability of the neural network, the robot can learn the mode in the high-dimensional space and the behavior track of the expert planner from the data. Thus, strategies (e.g., sampling mechanisms) in the planning process can be optimized to achieve excellent performance in a high-dimensional continuous space. Learning-based methods treat motion planning as a sequence decision problem, which can be generally solved by reinforcement learning. Previous work has proposed various neural network-based motion planners, including convolutional, recurrent, and graph neural network-based planners.
GNN-based planners exhibit significant capabilities in handling high-dimensional continuous planning tasks, such as GNN-Explorer and GraphMP. By accessing random geometry maps formed by uniform sampling, these planners are able to capture the geometric patterns of the environment, thereby enabling more efficient exploration without encoding the entire workspace. Although the GNN-based planner is excellent in high-dimensional continuous tasks, there are some limitations in that when an initial random geometry map (Random Geometric Graph, RGG) is constructed, the process of uniformly sampling the obtained nodes ignores the environment topology, which may lead to unnecessary node exploration, in the edge formation process of the RGG, edges colliding with obstacles may exist, thereby reducing the reliability of planning, and when a key map mode is extracted from a model, the existing method mainly encodes the positional relationship of the nodes, ignores the structural relationship among the nodes, and hinders the improvement of the expression capability of the model.
In the prior art, an efficient and accurate robot motion planning method based on human brain space memory heuristic is lacking.
Disclosure of Invention
In order to solve the technical problems of low node exploration efficiency and poor reliability of motion route planning in the prior art, the embodiment of the invention provides a method and a device for planning the motion of a robot inspired by spatial relationship memory. The technical scheme is as follows:
in one aspect, a method for planning motion of a robot inspired by spatial relationship memory is provided, the method being implemented by a robot motion planning device, the method comprising:
Acquiring movement starting position information, movement target position information, barrier position information and map information;
based on a selective sampling strategy, constructing a first random geometric figure in a configuration space of the robot through a K-NN algorithm according to the motion initial position information, the motion target position information and the map information;
inputting the first random geometric figure and the obstacle position information into the intelligent edge selector for edge screening based on a preset confidence threshold value to obtain a second random geometric figure;
According to the second random geometric figure, heuristic value calculation is carried out through the memory perception predictor, and node heuristic values are obtained;
Based on microscopic features And the module is used for searching the optimal path by using a quick path exploration method according to the node heuristic value and the second random geometric figure to obtain a second optimal path.
In another aspect, a robot motion planning apparatus inspired by spatial relationship memory is provided, the apparatus is applied to a robot motion planning method inspired by spatial relationship memory, the apparatus includes:
the information acquisition module is used for acquiring movement starting position information, movement target position information, barrier position information and map information;
The first random geometric figure constructing module is used for constructing a first random geometric figure through a K-NN algorithm in a configuration space of the robot according to the motion initial position information, the motion target position information and the map information based on a selective sampling strategy;
the second random geometric figure construction module is used for inputting the first random geometric figure and the obstacle position information into the intelligent edge selector for edge screening based on a preset confidence coefficient threshold value to obtain a second random geometric figure;
the heuristic value calculation module is used for calculating a heuristic value through the memory perception predictor according to the second random geometric figure to obtain a node heuristic value;
An optimal path acquisition module for based on microminiaturizable And the module is used for searching the optimal path by using a quick path exploration method according to the node heuristic value and the second random geometric figure to obtain a second optimal path.
In another aspect, a robotic motion planning apparatus is provided that includes a processor and a memory having stored thereon computer readable instructions that, when executed by the processor, implement any one of the above-described spatial relationship memory-inspired robotic motion planning methods.
In another aspect, a computer readable storage medium having stored therein at least one instruction loaded and executed by a processor to implement any of the above-described spatial relationship memory-inspired robot motion planning methods is provided.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
The invention provides a robot motion planning method inspired by spatial relation memory. In the process of decision making and relationship reasoning of human brain, the brain mechanism related to spatial relationship memory presents obvious advantages. Based on the method, a new brain-like motion planning based on a graph neural network is designed to improve the performance of robot planning in a high-dimensional continuous environment. When the random geometric figure is constructed, the sampling node is closer to the optimal solution through a new random geometric figure generating method, the searching efficiency is improved, the edge with the highest credibility in the random geometric figure is reserved through an intelligent edge selector, the negative influence of unreasonable edges on motion planning is avoided, the sensory observation information and the relation structure are integrated into a sensory reasoning brain mechanism through a memory perception predictor, the sensory observation information and the relation structure are simulated by a memory perception predictor, the graph representation capability of a graph neural network is improved by introducing the sub-graph structure into the node characteristic, and the integral learning effect of the model is enhanced. The invention relates to a high-efficiency and accurate robot motion planning method inspired by the memory of the brain space relation.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a robot motion planning method inspired by spatial relationship memory according to an embodiment of the invention;
FIG. 2 is a schematic diagram of an intelligent edge selector according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a memory perception predictor according to an embodiment of the present invention;
FIG. 4 is a block diagram of a robot motion planning apparatus inspired by spatial relationship memory according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a robot motion planning apparatus according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is described below with reference to the accompanying drawings.
In embodiments of the invention, words such as "exemplary," "such as" and the like are used to mean serving as an example, instance, or illustration. Any embodiment or design described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments or designs. Rather, the term use of an example is intended to present concepts in a concrete fashion. Furthermore, in embodiments of the present invention, the meaning of "and/or" may be that of both, or may be that of either, optionally one of both.
In the embodiments of the present invention, "image" and "picture" may be sometimes used in combination, and it should be noted that the meaning of the expression is consistent when the distinction is not emphasized. "of", "corresponding (corresponding, relevant)" and "corresponding (corresponding)" are sometimes used in combination, and it should be noted that the meaning of the expression is consistent when the distinction is not emphasized.
In embodiments of the present invention, sometimes a subscript such as W 1 may be written in a non-subscript form such as W1, and the meaning of the expression is consistent when de-emphasizing the distinction.
In order to make the technical problems, technical solutions and advantages to be solved more apparent, the following detailed description will be given with reference to the accompanying drawings and specific embodiments.
The embodiment of the invention provides a robot motion planning method inspired by spatial relation memory, which can be realized by a robot motion planning device, wherein the robot motion planning device can be a terminal or a server. The robot motion planning method flow diagram inspired by the spatial relationship memory as shown in fig. 1, the process flow of the method can comprise the following steps:
S1, acquiring movement starting position information, movement target position information, barrier position information and map information.
In a possible implementation, in a robot high-dimensional continuous motion scene, position information of a starting point and a target is obtained from information released by a ROS system of the robot. The robot-mounted laser sensor performs field scanning to acquire obstacle position information and current map information.
S2, based on a selective sampling strategy, constructing a first random geometric figure in a configuration space of the robot through a K-NN algorithm according to the motion initial position information, the motion target position information and the map information.
Optionally, based on a selective sampling strategy, according to the motion start position information, the motion target position information and the map information, constructing a first random geometric figure in a configuration space of the robot through a K-NN algorithm, wherein the method comprises the following steps:
Performing robot position mapping according to the movement initial position information, the movement target position information and the map information to obtain robot positioning information;
based on the configuration space of the robot, constructing a node set by using a selective sampling strategy according to the motion starting position information, the map information and the robot positioning information;
according to the node set, searching adjacent nodes by using a K-NN algorithm to obtain the adjacent node set;
and performing edge connection according to the node set and the adjacent node set to obtain a first random geometric figure.
In a possible implementation, the robot positioning information of the robot is determined according to the accurate spatial position description of the obstacle in the environment where the robot is located and matching the accurate spatial position description with a preset map.
In the planning process, the method aims to solve the problem that the random sampling ignores environmental information when constructing RGG. A batch of nodes is generated by the robot sampling with alpha probability in the region between the origin and the target, while sampling uniformly throughout the configuration space with (1-alpha) probability. The method can flexibly switch between biased sampling and uniform sampling by adjusting the super parameter alpha, balance efficiency and exploration, maintain probability completeness and reduce exploration of redundant nodes.
The invention adopts K-nearest neighbor algorithm (K-Nearest Neighbors algorithm, K-NN) algorithm to search adjacent nodes to construct a first random geometric figure. The first random geometry gives a feasible path of the robot in the configuration space. By this figure, a collision-free path from the initial configuration to the target configuration can be efficiently planned.
S3, inputting the first random geometric figure and the obstacle position information into an intelligent edge selector for edge screening based on a preset confidence threshold value, and obtaining a second random geometric figure.
Wherein the intelligent edge selector comprises a coding layer, a transducer layer and a hiding layer;
the coding layer is constructed according to the attention mechanism;
The hidden layer is constructed from a multi-layer perceptron model structure.
In a possible implementation manner, the architecture of the intelligent edge selector is shown in fig. 2, the coding layer comprises three encoders for respectively extracting features of nodes and edges of the first random geometric figure and positions of obstacles, the transducer layer comprises two transducer modules, and the hidden layer comprises three multi-layer sensors.
The intelligent edge selector is used in training learning stage, and each planning problem in the training data is composed of obstacle and initial vertexAnd a target vertexComposition is prepared. Construction of k-NN graphs by uniformly sampling nodes in a configuration spaceWherein. Loss by Binary Cross Entropy (BCE)Training to minimize confidence differences between the estimated value and the real label. Dijkstra algorithm as expert planner to provide real labels for training problems。
Optionally, inputting the first random geometry and the obstacle position information into the intelligent edge selector for edge screening based on a preset confidence threshold, to obtain a second random geometry, including:
Obtaining node information and side information according to the first random geometric figure;
Inputting the node information, the side information and the barrier position information into a coding layer for feature extraction to obtain a first node feature, a first side feature and a barrier position feature;
according to the first node characteristics, the first edge characteristics and the barrier position characteristics, performing coding processing through a transducer layer to obtain node-barrier characteristics and edge-barrier characteristics;
inputting the node-barrier characteristics and the edge-barrier characteristics into a hidden layer for confidence calculation to obtain edge confidence;
and screening the edges of the first random geometric figure according to the edge confidence degree and a preset confidence degree threshold value to obtain a second random geometric figure.
In a possible embodiment, the first random geometry RGGAs input to the encoder, the vertices and edges are encoded into the potential space. Each vertexAnd each edgeIs expressed as (1)AndWhereinAndIs two different two-layer MLP. In iterative obstacle codingThe obstacle encoding process in the t-th iteration of (c) may be expressed as the following equation (1):
(1)
Wherein, AndRespectively outputting the vertex and the edge after the barrier codes are iterated t-1 times,The representation layer is normalized, B is the output of the obstacle after passing through the encoder, Is an MLP layer. Output of the vertex and the edge after iterating t times of obstacle codes respectively, wherein、Is that、Intermediate results are output through the transducer layer in the iterative updating process. Is a typical attention function in whichRepresenting keys, queries and values, respectively.
RGGEach edge of (a)Can be evaluated using the obstacle information. Specifically, obstacle encoder with T iterationsInformation to obtain RGG and obstacles is encoded as each edgeIs expressed as . Edge(s)Through a three-layer MLP with two-dimensional outputCalculation, expressed as. Preserving edges with confidence higher than a preset threshold value to obtain a second random geometric figure。
And S4, according to the second random geometric figure, performing heuristic value calculation through a memory perception predictor to obtain a node heuristic value.
The memory perception predictor comprises a subgraph extraction layer, a graph coding layer, a linear layer, a pooling layer and an output layer.
In a possible implementation manner, the memory perception predictor provided by the invention is constructed based on a graph neural network, as shown in fig. 3. The extraction of the Ego subgraph and the Cut subgraph is included in the subgraph extraction layer, and the feature extraction of the vertex and the edge of the second random geometric graph is included. The figure coding layer comprises three encoders respectively corresponding to Ego subgraph, cut subgraph, second most-contraindicated geometric figure vertex and edge feature codes. The output layer is built up of three multi-layer perceptrons.
The memory sensor can be made microThe module, the memory-aware predictor, can be trained jointly in an end-to-end fashion. During the learning phase, training data is searched for problem instances by graphComposition of whichAll nodes and edges in the input RGG respectively,AndThe starting vertex and the target vertex are respectively,Is a binary vector, marking the vertices in the optimal path as 1. Training loss is defined as follows (2):
(2)
Wherein, Is made of a microminiaturizable materialThe closed list vector estimated by the module,Is the binary vector provided by the expert planner for the ith training problem.
Training loss is guided towards the optimal path by punishing the over-explored vertices in the path. Thus, by minimizing the training function, the training function can be made very smallThe optimal path can be searched efficiently.
Optionally, according to the second random geometric figure, performing heuristic value calculation through the memory perception predictor to obtain a node heuristic value, including:
inputting the second random geometric figure into a subgraph extraction layer to extract subgraphs to obtain an Ego subgraph and a Cut subgraph;
based on a preset random walk process, probability distribution calculation is carried out according to the Ego subgraph and the Cut subgraph, so that Ego subgraph random walk probability and Cut subgraph random walk probability are obtained;
inputting the second random geometric figure into a figure coding layer for feature extraction to obtain a second node feature and a second edge feature;
inputting the Ego sub-graph random walk probability and the Cut sub-graph random walk probability into a graph coding layer for coding processing to obtain a first hidden representation vector feature and a second hidden representation vector feature;
Splicing the first hidden representation vector feature and the second hidden representation vector feature to the vertex vector through the linear layer to obtain a first spliced vertex vector;
Inputting the first spliced vertex vector into a pooling layer for aggregation iteration to obtain a second spliced vertex vector;
And performing heuristic value calculation on the second spliced vertex vector and the second edge feature vector input-output layer to obtain node heuristic values of the second random geometric figure.
In one possible implementation, the present invention employs two sub-graph extraction strategies, ego subgraph and Cut subgraph. Ego subgraphDefining a k-hop EgoNetwork with node v as root, including neighbor nodes in k-hop of node vWherein, the method comprises the steps of, wherein,Representing a set of nodes within k-hop from the root node v. The Cut subgraph obtains a plurality of Cut subgraphs from the original graph by continuously and selectively removing edges.
Extracted subgraphThe ith vertex of (b)The random walk return probability coding of (c) is defined as follows (3):
(3)
Wherein, Representing sub-graphsMiddle root vertexReturn probability of the s-step random walk of departure. The random walk return probabilities of Ego and Cut subgraphs are expressed asAnd. Subsequently, two linear layers are applied to encode the random walk return probabilities of the Ego subgraph and the Cut subgraph into hidden representation vectors, respectivelyAnd。
Given a givenAnd a target vertexVertex thenFeatures of (2)EdgeFeatures of (2)Is defined as the following formula (4):
(4)
Wherein, AndIs two different two-layer MLP.
Ego subgraph and Cut subgraph are mapped to AND through two different two-layer MLPsThe same potential space. Then, the hidden representations of the two types of subgraphs are spliced with the vertex characteristics to obtain subgraph information injection characteristics, namelyAndThe following formula (5) is defined:
(5)
The neural network is constructed by combining the neighbors from each vertex Aggregating local information, iteratively updating Ego subgraph and Cut subgraph information to update node characteristics. The update mode of the Ego message transmission channel is as follows (6):
(6)
Wherein, AndIs three different two-layer MLP. The Cut messaging channel is identical to the Ego channel.
After L iterations, ego subgraph and Cut subgraph information injection features are respectively expressed asAnd. Finally, obtaining the complete graph through the pooling operation represents the following formula (7)
(7)
Wherein the POOL operation is a global pooling function for aggregating information of all vertices.
Vertex pointThe perceptual prediction of (1) is calculated as H,Is a three-layer MLP.
S5, based on microminiaturizationAnd the module is used for searching the optimal path by using a quick path exploration method according to the node heuristic value and the second random geometric figure to obtain a second optimal path.
Alternatively, based on a microminiaturizable basisThe module performs optimal path searching by using a quick path searching method according to the node heuristic value and the second random geometric figure to obtain a second optimal path, wherein the second optimal path comprises
Inputting node heuristic values can be microThe module is used for searching an optimal path in the second random geometric figure to obtain a first optimal path;
And deleting redundant nodes according to the first optimal path by a shortcut path exploration method to obtain a second optimal path.
In a possible implementation, the heuristic value is taken as a differentiable valueThe input of the module performs an optimal path search in the second random geometry RGG'.
Can be made microThe module performs the expansion process of the new node as follows (8):
(8)
Wherein the method comprises the steps of Representing the selected node, initial open listRepresenting all nodes of the random geometry map,Representing the total accumulated path cost of the starting point from the current nodeThe estimated cost of the current node from the endpoint is represented and replaced by a heuristic H. Gradually follow the selected node with the progress of searchingDelete and willStored in the closed list C.
In the micro AAfter the module completes the path search, the path quality still needs to be improved because some detours may exist. The present invention therefore contemplates a quick path exploration to identify potential shorter sides. Once these shortcuts are identified, they will be added to the exploration tree to eliminate unnecessary detours.
The search window length is preset to 2. Assuming that the current window sequentially contains the nodes v i-1, vi, vi+1, judging whether a shortcut exists between the nodes v i-1 and v i+1. If the edge between v i-1 and v i+1 does not intersect an obstacle, a shortcut exists within the search window. Once these shortcuts are identified, they will be added to the exploration tree to eliminate unnecessary detours. And iteratively checking whether a shortcut exists in the window by moving the search window so as to improve the path quality.
In a possible embodiment, the present invention uses five key metrics to evaluate performance, namely the success rate of finding collision-free paths (Success Rate of Collision-FREE PATHFINDING, SR), the number of collision detections (Number of Collision Checks, CK), the planning time (PLANNING TIME, PT), the Path Cost (Path Cost, CO), and the Path Cost (Penalized Path Cost, PO) that introduces penalty terms.
The performance comparisons of the model and the most advanced method proposed by the present invention on the above 6 data sets are shown in table 1 (planning performance comparison table on maze planning data set) and table 2 (planning performance comparison on robot arm operation data set).
TABLE 1
TABLE 2
The invention has obvious performance in all the tested simulation environments, and achieves the success rate close to 100%. Particularly in a high-dimensional environment, the invention greatly reduces the number of collision checks and the path cost. For some methods with lower success rates, it can only successfully find paths in the simpler example problem, with lower path costs. For this reason, the present invention introduces a penalty term in the failed path to balance the fairness of path cost comparisons. For this index, the present invention achieves optimal results in all simulation environments. Experimental results show that the method has robustness and stability for decision tasks in various complex environments.
The invention provides a robot motion planning method inspired by spatial relation memory. In the process of decision making and relationship reasoning of human brain, the brain mechanism related to spatial relationship memory presents obvious advantages. Based on the method, a new brain-like motion planning based on a graph neural network is designed to improve the performance of robot planning in a high-dimensional continuous environment. When the random geometric figure is constructed, the sampling node is closer to the optimal solution through a new random geometric figure generating method, the searching efficiency is improved, the edge with the highest credibility in the random geometric figure is reserved through an intelligent edge selector, the negative influence of unreasonable edges on motion planning is avoided, the sensory observation information and the relation structure are integrated into a sensory reasoning brain mechanism through a memory perception predictor, the sensory observation information and the relation structure are simulated by a memory perception predictor, the graph representation capability of a graph neural network is improved by introducing the sub-graph structure into the node characteristic, and the integral learning effect of the model is enhanced. The invention relates to a high-efficiency and accurate robot motion planning method inspired by the memory of the brain space relation.
FIG. 4 is a block diagram illustrating a spatial relationship memory inspired robot motion planning apparatus for a spatial relationship memory inspired robot motion planning method, according to an example embodiment. Referring to fig. 4, the apparatus includes an information acquisition module 410, a first random geometry construction module 420, a second random geometry construction module 430, a heuristic calculation module 440, and an optimal path acquisition module 450. Wherein:
an information acquisition module 410 for acquiring movement start position information, obstacle position information, and map information;
The first random geometry map construction module 420 is configured to construct a first random geometry map in a configuration space of the robot according to the motion start position information, the motion target position information and the map information based on a selective sampling strategy by a K-NN algorithm;
The second random geometry map construction module 430 is configured to input the first random geometry map and the obstacle location information into the intelligent edge selector for edge screening based on a preset confidence threshold value, so as to obtain a second random geometry map;
the heuristic value calculation module 440 is configured to perform heuristic value calculation by using the memory perception predictor according to the second random geometric figure, so as to obtain a node heuristic value;
An optimal path acquisition module 450 for based on the microminiaturizable And the module is used for searching the optimal path by using a quick path exploration method according to the graph node heuristic value and the second random geometric graph to obtain a second optimal path.
Optionally, the first random geometry building module 420 is further configured to:
Performing robot position mapping according to the movement initial position information, the movement target position information and the map information to obtain robot positioning information;
based on the configuration space of the robot, constructing a node set by using a selective sampling strategy according to the motion starting position information, the map information and the robot positioning information;
according to the node set, searching adjacent nodes by using a K-NN algorithm to obtain the adjacent node set;
and performing edge connection according to the node set and the adjacent node set to obtain a first random geometric figure.
Wherein the intelligent edge selector comprises a coding layer, a transducer layer and a hiding layer;
the coding layer is constructed according to the attention mechanism;
The hidden layer is constructed from a multi-layer perceptron model structure.
Optionally, the second random geometry building block 430 is further configured to:
Obtaining node information and side information according to the first random geometric figure;
Inputting the node information, the side information and the barrier position information into a coding layer for feature extraction to obtain a first node feature, a first side feature and a barrier position feature;
according to the first node characteristics, the first edge characteristics and the barrier position characteristics, performing coding processing through a transducer layer to obtain node-barrier characteristics and edge-barrier characteristics;
inputting the node-barrier characteristics and the edge-barrier characteristics into a hidden layer for confidence calculation to obtain edge confidence;
and screening the edges of the first random geometric figure according to the edge confidence degree and a preset confidence degree threshold value to obtain a second random geometric figure.
The memory perception predictor comprises a subgraph extraction layer, a graph coding layer, a linear layer, a pooling layer and an output layer.
Optionally, the heuristic value calculation module 440 is further configured to:
inputting the second random geometric figure into a subgraph extraction layer to extract subgraphs to obtain an Ego subgraph and a Cut subgraph;
based on a preset random walk process, probability distribution calculation is carried out according to the Ego subgraph and the Cut subgraph, so that Ego subgraph random walk probability and Cut subgraph random walk probability are obtained;
inputting the second random geometric figure into a figure coding layer for feature extraction to obtain a second node feature and a second edge feature;
inputting the Ego sub-graph random walk probability and the Cut sub-graph random walk probability into a graph coding layer for coding processing to obtain a first hidden representation vector feature and a second hidden representation vector feature;
Splicing the first hidden representation vector feature and the second hidden representation vector feature to the vertex vector through the linear layer to obtain a first spliced vertex vector;
Inputting the first spliced vertex vector into a pooling layer for aggregation iteration to obtain a second spliced vertex vector;
And performing heuristic value calculation on the second spliced vertex vector and the second edge feature vector input-output layer to obtain node heuristic values of the second random geometric figure.
Optionally, the optimal path acquisition module 450 is further configured to
Inputting graph node heuristic values can be microThe module is used for searching an optimal path in the second random geometric figure to obtain a first optimal path;
And deleting redundant nodes according to the first optimal path by a shortcut path exploration method to obtain a second optimal path.
The invention provides a robot motion planning method inspired by spatial relation memory. In the process of decision making and relationship reasoning of human brain, the brain mechanism related to spatial relationship memory presents obvious advantages. Based on the method, a new brain-like motion planning based on a graph neural network is designed to improve the performance of robot planning in a high-dimensional continuous environment. When the random geometric figure is constructed, the sampling node is closer to the optimal solution through a new random geometric figure generating method, the searching efficiency is improved, the edge with the highest credibility in the random geometric figure is reserved through an intelligent edge selector, the negative influence of unreasonable edges on motion planning is avoided, the sensory observation information and the relation structure are integrated into a sensory reasoning brain mechanism through a memory perception predictor, the sensory observation information and the relation structure are simulated by a memory perception predictor, the graph representation capability of a graph neural network is improved by introducing the sub-graph structure into the node characteristic, and the integral learning effect of the model is enhanced. The invention relates to a high-efficiency and accurate robot motion planning method inspired by the memory of the brain space relation.
Fig. 5 is a schematic structural diagram of a robot motion planning apparatus according to an embodiment of the present invention, where, as shown in fig. 5, the robot motion planning apparatus may include a robot motion planning device that is inspired by the spatial relationship memory shown in fig. 4. Optionally, the robotic motion planning device 510 may comprise a first processor 2001.
Optionally, the robotic motion planning device 510 may also include a memory 2002 and a transceiver 2003.
The first processor 2001 may be connected to the memory 2002 and the transceiver 2003, for example, via a communication bus.
The following describes the respective constituent elements of the robot motion planning apparatus 510 in detail with reference to fig. 5:
The first processor 2001 is a control center of the robot motion planning apparatus 510, and may be one processor or a plurality of processing elements. For example, the first processor 2001 is one or more central processing units (central processing unit, CPU), or may be an Application SPECIFIC INTEGRATED Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more microprocessors (DIGITAL SIGNAL processors, DSPs), or one or more field programmable gate arrays (field programmable GATE ARRAY, FPGAs).
Alternatively, the first processor 2001 may perform various functions of the robotic motion planning device 510 by running or executing a software program stored in the memory 2002, and invoking data stored in the memory 2002.
In a specific implementation, first processor 2001 may include one or more CPUs, such as CPU0 and CPU1 shown in fig. 5, as an example.
In a specific implementation, as an embodiment, the robotic motion planning device 510 may also include a plurality of processors, such as the first processor 2001 and the second processor 2004 shown in fig. 5. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor herein may refer to one or more devices, circuits, and/or processing cores for processing data (e.g., computer program instructions).
The memory 2002 is used for storing a software program for executing the solution of the present invention, and is controlled by the first processor 2001 to execute the solution, and the specific implementation may refer to the above method embodiment, which is not described herein.
Alternatively, memory 2002 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (random access memory, RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (ELECTRICALLY ERASABLE PROGRAMMABLE READ-only memory, EEPROM), compact disc read-only memory (compact disc read-only memory) or other optical disk storage, optical disk storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, without limitation. The memory 2002 may be integrated with the first processor 2001, may be present separately, and may be coupled to the first processor 2001 via an interface circuit (not shown in fig. 5) of the robotic motion planning device 510, as embodiments of the invention are not specifically limited in this regard.
A transceiver 2003 for communicating with a network device or with a terminal device.
Alternatively, transceiver 2003 may include a receiver and a transmitter (not separately shown in fig. 5). The receiver is used for realizing the receiving function, and the transmitter is used for realizing the transmitting function.
Alternatively, the transceiver 2003 may be integrated with the first processor 2001, or may exist separately, and be coupled to the first processor 2001 through an interface circuit (not shown in fig. 5) of the robotic motion planning device 510, as embodiments of the invention are not specifically limited in this regard.
It should be noted that the configuration of the robotic motion planning device 510 shown in fig. 5 is not limiting of the router, and that an actual knowledge structure recognition device may include more or fewer components than shown, or may combine certain components, or may have a different arrangement of components.
In addition, the technical effects of the robot motion planning apparatus 510 may refer to the technical effects of the robot motion planning method inspired by the spatial relationship memory described in the above method embodiments, which are not described herein.
It is to be appreciated that the first processor 2001 in embodiments of the invention may be a central processing unit (central processing unit, CPU) which may also be other general purpose processors, digital signal processors (DIGITAL SIGNAL processors, DSPs), application Specific Integrated Circuits (ASICs), off-the-shelf programmable gate arrays (field programmable GATE ARRAY, FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
It should also be appreciated that the memory in embodiments of the present invention may be either volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. The nonvolatile memory may be a read-only memory (ROM), a Programmable ROM (PROM), an erasable programmable ROM (erasable PROM), an electrically erasable programmable EPROM (EEPROM), or a flash memory. The volatile memory may be random access memory (random access memory, RAM) which acts as external cache memory. By way of example, and not limitation, many forms of random access memory (random access memory, RAM) are available, such as static random access memory (STATIC RAM, SRAM), dynamic Random Access Memory (DRAM), synchronous Dynamic Random Access Memory (SDRAM), double data rate synchronous dynamic random access memory (double DATA RATE SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (ENHANCED SDRAM, ESDRAM), synchronous link dynamic random access memory (SYNCHLINK DRAM, SLDRAM), and direct memory bus random access memory (direct rambus RAM, DR RAM).
The above embodiments may be implemented in whole or in part by software, hardware (e.g., circuitry), firmware, or any other combination. When implemented in software, the above-described embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with embodiments of the present invention are produced in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains one or more sets of available media. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium may be a solid state disk.
It should be understood that the term "and/or" is merely an association relationship describing the associated object, and means that three relationships may exist, for example, a and/or B, and may mean that a exists alone, while a and B exist alone, and B exists alone, wherein a and B may be singular or plural. In addition, the character "/" herein generally indicates that the associated object is an "or" relationship, but may also indicate an "and/or" relationship, and may be understood by referring to the context.
In the present invention, "at least one" means one or more, and "a plurality" means two or more. "at least one of" or the like means any combination of these items, including any combination of single item(s) or plural items(s). For example, at least one (a, b, or c) of a, b, c, a-b, a-c, b-c, or a-b-c may be represented, wherein a, b, c may be single or plural.
It should be understood that, in various embodiments of the present invention, the sequence numbers of the foregoing processes do not mean the order of execution, and the order of execution of the processes should be determined by the functions and internal logic thereof, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
It will be clearly understood by those skilled in the art that, for convenience and brevity of description, specific working procedures of the apparatus, device and unit described above may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
In the several embodiments provided by the present invention, it should be understood that the disclosed apparatus, device and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another device, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. The storage medium includes a U disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, an optical disk, or other various media capable of storing program codes.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A method for planning motion of a robot inspired by spatial relationship memory, the method comprising:
Acquiring movement starting position information, movement target position information, barrier position information and map information;
based on a selective sampling strategy, constructing a first random geometric figure in a configuration space of the robot through a K-NN algorithm according to the motion initial position information, the motion target position information and the map information;
inputting the first random geometric figure and the obstacle position information into the intelligent edge selector for edge screening based on a preset confidence threshold value to obtain a second random geometric figure;
According to the second random geometric figure, heuristic value calculation is carried out through the memory perception predictor, and node heuristic values are obtained;
Based on microscopic features And the module is used for searching the optimal path by using a quick path exploration method according to the node heuristic value and the second random geometric figure to obtain a second optimal path.
2. The robot motion planning method according to claim 1, wherein the constructing a first random geometry by a K-NN algorithm in a configuration space of the robot according to the motion start position information, the motion target position information, and the map information based on the selective sampling strategy comprises:
performing robot position mapping according to the motion initial position information, the motion target position information and the map information to obtain robot positioning information;
Based on the configuration space of the robot, constructing a node set by using a selective sampling strategy according to the motion starting position information, the map information and the robot positioning information;
according to the node set, performing adjacent node search by using a K-NN algorithm to obtain an adjacent node set;
and performing edge connection according to the node set and the adjacent node set to obtain a first random geometric figure.
3. The method of claim 1, wherein the intelligent edge selector comprises a coding layer, a transducer layer, and a hidden layer;
The coding layer is constructed according to an attention mechanism;
the hidden layer is constructed according to a multi-layer perceptron model structure.
4. The method for planning motion of a robot inspired by spatial relationship memory according to claim 3, wherein inputting the first random geometry and the obstacle position information into the intelligent edge selector for edge screening based on a preset confidence threshold to obtain a second random geometry comprises:
obtaining node information and side information according to the first random geometric figure;
inputting the node information, the side information and the barrier position information into the coding layer for feature extraction to obtain a first node feature, a first side feature and a barrier position feature;
According to the first node characteristics, the first edge characteristics and the barrier position characteristics, performing coding processing through the transformer layer to obtain node-barrier characteristics and edge-barrier characteristics;
inputting the node-obstacle characteristics and the edge-obstacle characteristics into the hidden layer for confidence calculation to obtain edge confidence;
And screening the edges of the first random geometric figure according to the edge confidence degree and a preset confidence degree threshold value to obtain a second random geometric figure.
5. The method of claim 1, wherein the memory aware predictor comprises a subgraph extraction layer, a graph coding layer, a linear layer, a pooling layer, and an output layer.
6. The method for planning motion of a robot inspired by spatial relationship memory according to claim 5, wherein the performing, according to the second random geometric figure, inspiring value calculation by the memory perception predictor, obtaining node inspiring value includes:
inputting the subgraph extraction layer according to the second random geometric figure to perform subgraph extraction to obtain an Ego subgraph and a Cut subgraph;
based on a preset random walk process, probability distribution calculation is carried out according to the Ego subgraph and the Cut subgraph, so that Ego subgraph random walk probability and Cut subgraph random walk probability are obtained;
inputting the second random geometric figure into the figure coding layer to perform feature extraction to obtain a second node feature and a second edge feature;
Inputting the Ego sub-graph random walk probability and the Cut sub-graph random walk probability into the graph coding layer for coding processing to obtain a first hidden representation vector feature and a second hidden representation vector feature;
Splicing the first hidden representation vector feature and the second hidden representation vector feature to the vertex vector through the linear layer to obtain a first spliced vertex vector;
Inputting the first spliced vertex vector into the pooling layer for aggregation iteration to obtain a second spliced vertex vector;
And inputting the second spliced vertex vector and the second edge feature vector into the output layer to perform heuristic value calculation, so as to obtain node heuristic values of a second random geometric figure.
7. The spatial relationship memory inspired robot motion planning method of claim 1, wherein the microminiaturizable basedAnd the module is used for searching the optimal path by using a quick path exploration method according to the node heuristic value and the second random geometric figure to obtain a second optimal path, and comprises the following steps:
inputting the node heuristic value into a microtopograph The module is used for searching an optimal path in the second random geometric figure to obtain a first optimal path;
And deleting redundant nodes according to the first optimal path by a shortcut path exploration method to obtain a second optimal path.
8. A spatial relationship memory inspired robot motion planning apparatus for implementing a spatial relationship memory inspired robot motion planning method according to any one of claims 1-7, characterized in that the apparatus comprises:
the information acquisition module is used for acquiring movement starting position information, movement target position information, barrier position information and map information;
The first random geometric figure constructing module is used for constructing a first random geometric figure through a K-NN algorithm in a configuration space of the robot according to the motion initial position information, the motion target position information and the map information based on a selective sampling strategy;
the second random geometric figure construction module is used for inputting the first random geometric figure and the obstacle position information into the intelligent edge selector for edge screening based on a preset confidence coefficient threshold value to obtain a second random geometric figure;
the heuristic value calculation module is used for calculating a heuristic value through the memory perception predictor according to the second random geometric figure to obtain a node heuristic value;
An optimal path acquisition module for based on microminiaturizable And the module is used for searching the optimal path by using a quick path exploration method according to the node heuristic value and the second random geometric figure to obtain a second optimal path.
9. A robotic motion planning apparatus, the robotic motion planning apparatus comprising:
a processor;
A memory having stored thereon computer readable instructions which, when executed by the processor, implement the method of any of claims 1 to 7.
10. A computer readable storage medium having stored therein program code which is callable by a processor to perform the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411193677.1A CN119124191B (en) | 2024-08-28 | 2024-08-28 | A robot motion planning method and device inspired by spatial relation memory |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411193677.1A CN119124191B (en) | 2024-08-28 | 2024-08-28 | A robot motion planning method and device inspired by spatial relation memory |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN119124191A true CN119124191A (en) | 2024-12-13 |
| CN119124191B CN119124191B (en) | 2025-09-30 |
Family
ID=93759609
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411193677.1A Active CN119124191B (en) | 2024-08-28 | 2024-08-28 | A robot motion planning method and device inspired by spatial relation memory |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119124191B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150339570A1 (en) * | 2014-05-22 | 2015-11-26 | Lee J. Scheffler | Methods and systems for neural and cognitive processing |
| WO2017214581A1 (en) * | 2016-06-10 | 2017-12-14 | Duke University | Motion planning for autonomous vehicles and reconfigurable motion planning processors |
| US20210179097A1 (en) * | 2019-12-12 | 2021-06-17 | Baidu Usa Llc | Lane-attention: predicting vehicles' moving trajectories by learning their attention over lanes |
| CN115705052A (en) * | 2021-08-10 | 2023-02-17 | 南方科技大学 | Robot motion path planning method and device, terminal equipment and medium |
| US20230206029A1 (en) * | 2021-12-27 | 2023-06-29 | International Business Machines Corporation | Graph Neural Network Ensemble Learning |
| CN116698017A (en) * | 2023-08-07 | 2023-09-05 | 中国科学院合肥物质科学研究院 | Object-level environment modeling method and system for indoor large-scale complex scene |
| CN117908542A (en) * | 2024-01-16 | 2024-04-19 | 西安交通大学 | Multi-agent path planning method and system based on logistics storage environment |
| CN118238129A (en) * | 2023-12-26 | 2024-06-25 | 北京师范大学 | Robot motion planning and obstacle avoidance decision method inspired by parallel reasoning mechanism |
| CN118445753A (en) * | 2024-04-30 | 2024-08-06 | 华中科技大学 | Underwater unmanned underwater vehicle track fusion method and system based on graph convolution neural network |
-
2024
- 2024-08-28 CN CN202411193677.1A patent/CN119124191B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150339570A1 (en) * | 2014-05-22 | 2015-11-26 | Lee J. Scheffler | Methods and systems for neural and cognitive processing |
| WO2017214581A1 (en) * | 2016-06-10 | 2017-12-14 | Duke University | Motion planning for autonomous vehicles and reconfigurable motion planning processors |
| US20210179097A1 (en) * | 2019-12-12 | 2021-06-17 | Baidu Usa Llc | Lane-attention: predicting vehicles' moving trajectories by learning their attention over lanes |
| CN115705052A (en) * | 2021-08-10 | 2023-02-17 | 南方科技大学 | Robot motion path planning method and device, terminal equipment and medium |
| US20230206029A1 (en) * | 2021-12-27 | 2023-06-29 | International Business Machines Corporation | Graph Neural Network Ensemble Learning |
| CN116698017A (en) * | 2023-08-07 | 2023-09-05 | 中国科学院合肥物质科学研究院 | Object-level environment modeling method and system for indoor large-scale complex scene |
| CN118238129A (en) * | 2023-12-26 | 2024-06-25 | 北京师范大学 | Robot motion planning and obstacle avoidance decision method inspired by parallel reasoning mechanism |
| CN117908542A (en) * | 2024-01-16 | 2024-04-19 | 西安交通大学 | Multi-agent path planning method and system based on logistics storage environment |
| CN118445753A (en) * | 2024-04-30 | 2024-08-06 | 华中科技大学 | Underwater unmanned underwater vehicle track fusion method and system based on graph convolution neural network |
Non-Patent Citations (2)
| Title |
|---|
| 姜英;王延江;林青;刘伟锋;: "基于卷积神经网络与Bayesian决策的图像识别与分类记忆建模", 中国科学:技术科学, no. 09, 30 September 2017 (2017-09-30) * |
| 王军平;张文生;王勇飞;孙正雅;: "面向大数据领域的事理认知图谱构建与推断分析", 中国科学:信息科学, no. 07, 31 December 2020 (2020-12-31) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN119124191B (en) | 2025-09-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tang et al. | # exploration: A study of count-based exploration for deep reinforcement learning | |
| Ding et al. | Predicting vehicle behaviors over an extended horizon using behavior interaction network | |
| Luo et al. | Jfp: Joint future prediction with interactive multi-agent modeling for autonomous driving | |
| CN115240786B (en) | Prediction method, training method, device and electronic device for reactant molecules | |
| CN115426671A (en) | Method, system and equipment for graph neural network training and wireless cell fault prediction | |
| Maglogiannis et al. | Artificial intelligence applications and innovations | |
| Chen et al. | Q‐EANet: Implicit social modeling for trajectory prediction via experience‐anchored queries | |
| CN109034020A (en) | A kind of community's Risk Monitoring and prevention method based on Internet of Things and deep learning | |
| CN118823731A (en) | A vehicle trajectory prediction method, model and electronic device based on improved Transformer model and target point guidance | |
| CN119963842B (en) | Three-dimensional point cloud semantic segmentation method based on graph convolution and group vector attention mechanism | |
| CN115034459A (en) | Pedestrian trajectory time sequence prediction method | |
| CN119513285A (en) | Vehicle trajectory prediction method based on large language model | |
| CN119389241A (en) | A method, system, storage medium and device for predicting automatic driving trajectory | |
| Lin et al. | OST-HGCN: optimized spatial-temporal hypergraph convolution network for trajectory prediction | |
| Chen et al. | Emexplorer: an episodic memory enhanced autonomous exploration strategy with voronoi domain conversion and invalid action masking | |
| Javaheripi et al. | Swann: Small-world architecture for fast convergence of neural networks | |
| CN119648749B (en) | Target tracking method and system based on space channel summation attention | |
| CN119124191B (en) | A robot motion planning method and device inspired by spatial relation memory | |
| US12393195B2 (en) | Method and system for reinforcement learning and dual channel action embedding based robotic navigation | |
| CN115963832B (en) | Autonomous Exploration Methods for Robots Based on Deep Reinforcement Learning | |
| Neubert et al. | A sequence-based neuronal model for mobile robot localization | |
| CN120760743A (en) | A robot motion planning method inspired by the human brain's situational semantic coordination mechanism | |
| Zou et al. | Fuseformer: A manifold metric fusing attention for pedestrian trajectory prediction | |
| Balasubramani et al. | Empowering Autonomous Decision-Making Through Quantum Reinforcement Learning and Cognitive Neuromorphic Frameworks | |
| Shilpashree et al. | Neuromorphic-Driven Agentic AI for Autonomous Decision-Making Systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |