Douglas Blank
Bryn Mawr College, Computer Science, Faculty Member
- Michael Gasseredit
Over the past 15 years our research group has been exploring models of developmental robotics and curiosity. Our research is based on the premise that intelligent behavior arises through emergent interactions between opposing forces in an... more
Over the past 15 years our research group has been exploring models of developmental robotics and curiosity. Our research is based on the premise that intelligent behavior arises through emergent interactions between opposing forces in an open-ended, task-independent environment. In an initial experiment we constructed a recurrent neural network model where self-motivation was "an emergent property generated by the competing pressures that arise in attempting to balance predictability and novelty" [5]. The system first focused on its error, then learned to successfully predict its error, and finally became habituated to what caused the error. This process of focusing, learning, and habituating can be seen as a rudimentary type of curiosity.
This paper describes and tests a developmental architecture that enables a robot to explore its world, to find and remember interesting states, to associate these states with grounded goal representations, and to generate action sequences... more
This paper describes and tests a developmental architecture that enables a robot to explore its world, to find and remember interesting states, to associate these states with grounded goal representations, and to generate action sequences so that it can re-visit these states of interest. The model is composed of feed-forward neural networks that learn to make predictions at two levels through a dual mechanism of motor babbling for discovering the interesting goal states and instant replay learning for developing the grounded goal representations. We compare the performance of the model with grounded goal representations versus random goal representations, and find that it is significantly better at re-visiting the goal states when using grounded goal representations.
We are largely in agreement with Tani's approach to developmental robotics as elucidated in this dialog and his recent book. The basic assumptions inherent in his approach, such as that agents are embodied in the world and that neural... more
We are largely in agreement with Tani's approach to developmental robotics as elucidated in this dialog and his recent book. The basic assumptions inherent in his approach, such as that agents are embodied in the world and that neural systems are capable of complex learning, are now established wisdom. Although this has been a relatively recent shift in AI and Cognitive Science, we consider these underlying assumptions to be a given and thus do not address them further. Here we expand on Tani's questions and offer a broader set of principles for guiding developmental robotics research.
Scene construction is the process of building realistic, three-dimensional representations, or models, of real world environments, such as rooms, landsacpes or buildings. Because of the realistic quality of images being produced, current... more
Scene construction is the process of building realistic, three-dimensional representations, or models, of real world environments, such as rooms, landsacpes or buildings. Because of the realistic quality of images being produced, current scene construction algorithms require manual processing by human experts. However, the benefits of having such 3D models are great. Consider a situation where a three-dimensional model of an evironment must be created in real-time. Existing scene construction algorithms will not su~ ce. Therefore we ...
This paper describes Pyro, a robotics programming environment designed to allow inexperienced undergraduates to explore topics in advanced robotics. Pyro, which stands for Python Robotics, runs on a number of advanced robotics platforms.... more
This paper describes Pyro, a robotics programming environment designed to allow inexperienced undergraduates to explore topics in advanced robotics. Pyro, which stands for Python Robotics, runs on a number of advanced robotics platforms. In addition, programs in Pyro can abstract away low-level details such that individual programs can work unchanged across very different robotics hardware. Results of using Pyro in
Research Interests:
Research Interests:
Research Interests:
Research Interests:
IntroductionCompetitions have been used to pit state-of-the-art robots against each other in recent yearswith great success in at least one area: this has been one of the most advertised events atmany AI meetings. However, besides the... more
IntroductionCompetitions have been used to pit state-of-the-art robots against each other in recent yearswith great success in at least one area: this has been one of the most advertised events atmany AI meetings. However, besides the hype, can contests actually be useful to furtherdrive developments in robotics? We believe that they can. In addition, we propose a modelof technology-sharing such that competitions could provide the support foundation. To thatend, we propose a set of recommendations for robotics competitions. 2 ProblemsEach ...
Summary: Pyro, which stands for Python Robotics, is a Python-based robotics programming environment that enables students to explore topics in robotics and artificial intelligence. Programming robot behaviors in Pyro is akin to... more
Summary: Pyro, which stands for Python Robotics, is a Python-based robotics programming environment that enables students to explore topics in robotics and artificial intelligence. Programming robot behaviors in Pyro is akin to programming in a high-level general purpose programming language; Pyro provides abstractions for low-level robot-specific features so that a single program can run on many different types of robots. The Pyro project consists of the software as well as a set of curriculum modules that can be assembled to ...
Scene construction is the process of building realistic, three-dimensional representations, or models, of real world environments, such as rooms, landsacpes or buildings. Because of the realistic quality of images being produced, current... more
Scene construction is the process of building realistic, three-dimensional representations, or models, of real world environments, such as rooms, landsacpes or buildings. Because of the realistic quality of images being produced, current scene construction algorithms require manual processing by human experts. However, the benefits of having such 3D models are great. Consider a situation where a three-dimensional model of an evironment must be created in real-time. Existing scene construction algorithms will not su~ ce. Therefore we ...
This thesis presents an innate behavior for a robot that is supposed to find an object and take it to a goal location. The specific behavior is designed for an Aibo robot to find a ball and kick it towards a goal until it scores. The... more
This thesis presents an innate behavior for a robot that is supposed to find an object and take it to a goal location. The specific behavior is designed for an Aibo robot to find a ball and kick it towards a goal until it scores. The behavior is implemented using a Finite State Machine and explores several aspects of robotics such as: object detection and recognition, navigation in an unknown environment, goal detection as well as goal achievement. Part of this thesis examines the design and implementation of a Python ...
Research Interests:
Over the past 15 years our research group has been exploring models of developmental robotics and curiosity. Our research is based on the premise that intelligent behavior arises through emergent interactions between opposing forces in an... more
Over the past 15 years our research group has been exploring
models of developmental robotics and curiosity. Our
research is based on the premise that intelligent behavior
arises through emergent interactions between opposing
forces in an open-ended, task-independent environment.
In an initial experiment we constructed a recurrent neural
network model where self-motivation was "an emergent
property generated by the competing pressures that arise
in attempting to balance predictability and novelty". The
system first focused on its error, then learned to successfully
predict its error, and finally became habituated to what
caused the error. This process of focusing, learning, and
habituating can be seen as a rudimentary type of curiosity.
models of developmental robotics and curiosity. Our
research is based on the premise that intelligent behavior
arises through emergent interactions between opposing
forces in an open-ended, task-independent environment.
In an initial experiment we constructed a recurrent neural
network model where self-motivation was "an emergent
property generated by the competing pressures that arise
in attempting to balance predictability and novelty". The
system first focused on its error, then learned to successfully
predict its error, and finally became habituated to what
caused the error. This process of focusing, learning, and
habituating can be seen as a rudimentary type of curiosity.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Abstract The governor architecture is a new method for avoiding catatrophic forgetting in neural networks that is particularly useful in online robot learning. The governor architecture uses a categorizer to identify events and excise... more
Abstract The governor architecture is a new method for avoiding catatrophic forgetting in neural networks that is particularly useful in online robot learning. The governor architecture uses a categorizer to identify events and excise long sequences of repetitive data that cause catastrophic forgetting in neural networks trained on robot-based tasks. We examine the performance of several variations of the governor architecture on a number of related localization tasks using a simulated robot.
Most everyone, including the experts, would agree that analogy-making is best defined as a process that creates a mapping between items in one domain (often called the source) to “similar” items in another domain (often called the... more
Most everyone, including the experts, would agree that analogy-making is best defined as a process that creates a mapping between items in one domain (often called the source) to “similar” items in another domain (often called the target). Based on this definition, many researchers have attempted to model analogy-making by creating a mapping between two sets of data structures that represent the domains (Gentner, 1983; Holyoak and Thagard, 1989).
2. INTRODUCTION As we move our classrooms to cyberspace, we find ourselves redefining many of the core concepts in education. Even the fundamental notion of “course” needs to be reexamined. Roger Schank notes that the length of time and... more
2. INTRODUCTION As we move our classrooms to cyberspace, we find ourselves redefining many of the core concepts in education. Even the fundamental notion of “course” needs to be reexamined. Roger Schank notes that the length of time and amount of material covered in a “course” is completely arbitrary and need not be reflected in Webbased versions [11]. Computer-based testing (CBT) promises to further stretch many traditional concepts, such as “exam'” and “homework.” And rightly so.
Most everyone, including the experts, would agree that analogy-making is best defined as a process that creates a mapping between items in one domain (often called the source) to" similar" items in another (often called the target).
Abstract We present a perspective on the design of a curriculum for a new computer science program at a women's libera! arts college. The design incorporates lessons learned at the college in its successful implementation of other... more
Abstract We present a perspective on the design of a curriculum for a new computer science program at a women's libera! arts college. The design incorporates lessons learned at the college in its successful implementation of other academic programs, incorporation of best practices in curriculum design at other colleges, results from studies performed on various computer science programs, and a significant number of our own ideas. Several observations and design decisions are presented as curriculum design patterns.
Abstract This paper explores a philosophy and connectionist algorithm for creating a long-term, self-motivated developmental robot control system. Self-motivation is viewed as an emergent property arising from two competing pressures: the... more
Abstract This paper explores a philosophy and connectionist algorithm for creating a long-term, self-motivated developmental robot control system. Self-motivation is viewed as an emergent property arising from two competing pressures: the need to accurately predict the environment while simultaneously wanting to seek out novelty in the environment. These competing internal pressures are designed to drive the system in a manner reminiscent of a co-evolutionary arms race.
It is difficult to clearly define the symbolic and subsymbolic paradigms; each is usually described by its tendencies rather than any one definitive property. Symbolic processing is generally characterized by hard-coded, explicit rules... more
It is difficult to clearly define the symbolic and subsymbolic paradigms; each is usually described by its tendencies rather than any one definitive property. Symbolic processing is generally characterized by hard-coded, explicit rules operating on discrete, static tokens, whereas subsymbolic processing is associated with learned, fuzzy constraints affecting continuous, distributed representations.
Abstract Anticipatory systems have been shown to be useful in discrete, symbolic systems. However, nonsymbolic anticipatory systems are less well understood. In this paper, we explore the use of anticipation within the framework of... more
Abstract Anticipatory systems have been shown to be useful in discrete, symbolic systems. However, nonsymbolic anticipatory systems are less well understood. In this paper, we explore the use of anticipation within the framework of connectionist networks to bootstrap from an innate behavior; to drive a reinforcement signal; and to provide feedback on the learnability of a task.
Abstract We describe myro. chuck, a Python module for controlling music synthesis, and its applications to teaching introductory computer science. The module was built within the Myro framework using the ChucK programming language, and... more
Abstract We describe myro. chuck, a Python module for controlling music synthesis, and its applications to teaching introductory computer science. The module was built within the Myro framework using the ChucK programming language, and was used in an introductory computer science course combining robots, graphics and music. The results supported the value of music in engaging students and broadening their view of computer science.
Abstract This dissertation explores the integration of learning and analogy-making through the development of a computer program, called Analogator, that learns to make analogies by example. By “seeing” many different analogy problems,... more
Abstract This dissertation explores the integration of learning and analogy-making through the development of a computer program, called Analogator, that learns to make analogies by example. By “seeing” many different analogy problems, along with possible solutions, Analogator gradually develops an ability to make new analogies. That is, it learns to make analogies by analogy.
Search and Rescue and Hors d'Oeuvres Anyone? events at the 2000 American Association of Artificial Intelligence autonomous robot competitions. This paper describes the multirobot, low-cost sound localization technique, and the... more
Search and Rescue and Hors d'Oeuvres Anyone? events at the 2000 American Association of Artificial Intelligence autonomous robot competitions. This paper describes the multirobot, low-cost sound localization technique, and the multisensor, person recognition system used in the AAAI contests by the robot team.
Abstract We use a connectionist network trained with reinforcement to control both an autonomous robot vehicle and a simulated robot. We show that given appropriate sensory data and architectural structure, a network can learn to control... more
Abstract We use a connectionist network trained with reinforcement to control both an autonomous robot vehicle and a simulated robot. We show that given appropriate sensory data and architectural structure, a network can learn to control the robot for a simple navigation problem. We then investigate a more complex goal-based problem and examine the plan-like behavior that emerges.
Shiffrin has made many contributions to the modeling of human cognition in areas ranging from perception to attention to learning, but is best known for his long-standing efforts to develop explicit models of human memory. His most recent... more
Shiffrin has made many contributions to the modeling of human cognition in areas ranging from perception to attention to learning, but is best known for his long-standing efforts to develop explicit models of human memory. His most recent models use Bayesian, adaptive approaches, building on previous work but extending it in a critical new manner, and carrying his theory beyond explicit memory to implicit learning and memory processes.
Abstract This work focuses on the role of self-motivation in the developmental learning process of a mobile robot. We are interested in developing a general learning architecture that will enable a robot to build up hierarchical... more
Abstract This work focuses on the role of self-motivation in the developmental learning process of a mobile robot. We are interested in developing a general learning architecture that will enable a robot to build up hierarchical representations of its experiences through the processes of abstraction, for example by learning topological maps of sensory states, and anticipation, in which the robot learns to predict the outcome of applying its effectors to its current situation.
Abstract The Association for the Advancement of Artificial Intelligence presented its 2005 Spring Symposium Series on Monday through Wednesday, March 21-23, 2005 at Stanford University in Stanford, California.
Abstract We present a cryptographic system based on Pollack's recursive auto-associative memory (1991). This self-organizing learning system offers many unique properties not found in traditional crypto systems. This paper examines the... more
Abstract We present a cryptographic system based on Pollack's recursive auto-associative memory (1991). This self-organizing learning system offers many unique properties not found in traditional crypto systems. This paper examines the methodology and issues of such a system.
Abstract A 5 degree-of-freedom binocular stereovision head, called Argus, is described which has been designed for independent camera control (other than that of occular vergence). The system was inspired by both the abilities and... more
Abstract A 5 degree-of-freedom binocular stereovision head, called Argus, is described which has been designed for independent camera control (other than that of occular vergence). The system was inspired by both the abilities and behaviors of the chameleon visual system. A unique control architecture is described which enables hierarchical learning of visual behaviors, including independent search for an object of interest coupled with dual-camera object fixation upon successful completion of the search.
Abstract This paper reports on a comparison to the well-known NetTalk implementation of Engl! sh text-to-speech translation via neural networks. A distributed representation scheme for encoding is investigated opposed to the classic... more
Abstract This paper reports on a comparison to the well-known NetTalk implementation of Engl! sh text-to-speech translation via neural networks. A distributed representation scheme for encoding is investigated opposed to the classic localist representation scheme used in the original NetTalk. The paper discusses a modem re-implementation based on Elman's Simple Recurrent Network.
In August 1998 Dave Touretzky asked on the connectionists e-mailing list," Is connectionist symbol processing dead?" This query lead to an interesting discussion and exchange of ideas. We thought it might be useful to capture this... more
In August 1998 Dave Touretzky asked on the connectionists e-mailing list," Is connectionist symbol processing dead?" This query lead to an interesting discussion and exchange of ideas. We thought it might be useful to capture this exchange in an article. We solicited contributions, and this collective article is the result. Contributions were solicited by a public call on the connectionists e-mailing list. All contributions received were subjected to two to three informal reviews. Almost all were accepted with varying degrees of revision.
Abstract For the past several years, we have been using robots in our introductory computer science course. Although this has been challenging for many reasons, it has also been very rewarding on a number of fronts, both for the students... more
Abstract For the past several years, we have been using robots in our introductory computer science course. Although this has been challenging for many reasons, it has also been very rewarding on a number of fronts, both for the students and for us. However, in order for this to occur, we had to adapt to what we perceived as “chaotic code.” In this paper we describe lessons learned by watching what the students do, where they have trouble, and what they enjoy.