CSC592: Adv Robot Programming Assignment #1 [Kastuv M.
Tuladhar 7411725]
1. How the navigation has been integrated?
Answer:
Navigation on the robot is dependent to how the robot interacts with human. For socially assistive robots, people tries
to assign human like characteristics so that human dont need to undergo the breakdown the human robot interaction
and the societal norms are kept intact. Computer vision and control algorithms are applied to model the robots
behavior based like human-human interaction. For instance, Nakauchi Simmons developed the algorithm for robot
stand in line using the personal space while Pacchierotti defines it on the basis of avoidance scenario using human-
human proxemics distances. The algorithms are implemented depending upon the human like their face detection,
head gestures and even emotional status, physical status, verbal communication etc.
By using Vision Algorithms and collecting information about the users attention, emotion and activity it can enables
the robot to analyze the non-verbal communication gesture of the user like the head gestures. Few Computer Vision
algorithms used are Human-robot localization, robot navigation, human-robot interaction using an online face
recognition and head pose estimation.
[Link] major differences between active and passive robots helps? Discuss with example(s).
Answer:
Active Robots generally provides the physical help with the provided input/command.
That includes:
Home electronics / actuators (turning lights on/off, turning motors on/off, vacuum cleaning)
Obeying commands to fetch newspaper, drinking water or washing dishes.
While the passive robots can provide help without the command or input.
That includes:
Using sensors to turn on light on dark.
Turning on water pumps when the reservoir reaches minimum threshold level etc.
Figure 1:Passive dynamically walking robot
CSC592: Adv Robot Programming Assignment #1 [Kastuv M. Tuladhar 7411725]
3. In the talk, it has been said that due to support vector machines computational complexity, neural
network was employed? Are you agreed with this? If so, to what extent?
Answer:
To some extent I agree with the author as there is risk of inaccurate samples to feed SVM the performance might
deteriorate. While NN Is the way of learning adaptively. It trains the robot just as our human brain. As we know that
the Neural network (NN) and Support Vector Machines (SVM) are two popular strategies for machine learning. Both
strategies have its own benefits depending upon the requirements.
The advantage of NN over SVM is that NN is parametric model i.e. fixed in size whereas SVM is non-parametric.
SVM consists of the support vectors that are selected from the training sets. Here the number of support vectors is
exactly the number of training samples. While SVM computation becomes complex with increase in support vectors,
NN misses the big picture as the learning parameters requires the time of the training. Furthermore, there is also risk
of overfitting i.e. if the training goes too long, it may presume the noise as a pattern.
4. a) How robust is the face detection algorithm is (with respect to the state-of-the-art algorithms)?
Answer:
The state of art vision technologies is the work combining localization, navigation, face detection and head pose
estimation algorithm for the use of assisted living home environment. The algorithm is able to detect a person robustly
by using various features like navigating the robot towards the person and maintain the eye contact and check if the
person is giving attention to robot.
b) If you were developing a computer vision tool for face detection (for the similar task), what features
you would have used except what the authors reported in their work? Enlist them (at least a few) with
clear scientific arguments.
Answer:
The other features that includes for face detection by combining different information and combine various visual
information to detect and localize the persons position. The algorithm is capable of tracking the person with or without
the motion information and it is robust against the environmental noise like moving furniture and changing light
condition.
CSC592: Adv Robot Programming Assignment #1 [Kastuv M. Tuladhar 7411725]
The work flow is divided into two parts: prediction and adaption
The prediction phase constitutes small particle segments patch and evaluate to visual cues. The color memory cues
based on color histogram, shape memory cue and fixed shape cue based on neural network. The Sigma-Pi model is
used to weights when multiple cues are active at the same time. The snapshot of history is maintained in order to
relatively compare the previous stages with the current images. The Gaussian Model can help to retrieve any wrong
position made by the robot. Thus, this method is also robust in comparison to other state of art algorithms.
5. What is the take home message from the paper and/or presentation?
Answer:
This paper gave us the brief idea about:
the different computer vision techniques used in robot development for Ambient Assisted Living(AAL).
We came to know about navigation implementation problem and some of the algorithm to solve the
navigation problem.
We got idea about different vision based tasks for simultaneous person and robot localization, robot-
navigation, human-robot interaction.
6. Can you rate (range: [1:5]) and suggest/comment on speakers
Answer:
a) presentation style, (4)
b) skill, and (4)
c) level of understanding? (4)
