Download Mobile Robots Learning-Fuzzy Logic-Lecture Handouts and more Lecture notes Artificial Intelligence in PDF only on Docsity!
Hiden behaviours f(s,z)a; Reactive control: sf(z,a) ; Blind motor behaviours : zf(z,a); State
dependent behaviours : f: (s,z) (a,z)
Mobile robots
In the previous section, a generic classification of PbD approaches is discussed. The focus of this thesis is
PbD for mobile robots, so a few applications of mobile robots are explicitly discussed in this section that
use PbD or similar learning techniques. The learning techniques for mobile robot applications could be
classified as follows:
begin{itemize}
\item Trajectory learning,
\item Reactive learning,
\item Learning to reach recognizable objects,
\item Learning to reach position targets.
end{itemize}
Trajectory learning
Trajectory learning approaches could further be divided into two classes as follows:
begin{itemize}
\item Memorizing a given trajectory and following it using a preprogrammed trajectory follower,
\item Learning a generalized and regenerative trajectory model.
end{itemize}
The first classes
presented a novel approach to classify different places in the environment into semantic classes, like rooms, hallways, corridors, and doorways. Our technique uses simple geometric features extracted
from a single laser \cite{ semanticMeaningofRangedata}
Learning to reach position targets (path planning)
A neurofuzzy controller is trained from demonstrations and then used to reach the target in a complex
environment. There could be situation of dead cycles. From demonstrations the reactive velocity and
acceleration plans are learned. Simulation are performed only. \cite{neurofuzzy}
A similar system is shown \cite{ immunological } with may trap conditions. Ultra sonic and camera are
used.
Learning trajectories
Memorizing the trajectories using Bazier Curves \cite{ bazierCurves }. Are used to instruct a trajectory on
a touch pad. A smooth trajectory is extracted from finger tip movement. Supervisoy control.
Trajectory memorizing is criticized. Regenerative trajectory learning techniques using dynamical system
modeling is proposed \cite{ ijspeert2001}
Trajectory learning and generalizing by linking subgoals to the objects in the environment (learning a
path like digit 8). Trajectory is segmented and modeled by linear Dynamical Systems \cite{
trajectorywithSubgoals }.
Trajectories modeled as LDS using range data, end point attractor and initial point. Attractor is
associated with the objects in environment to adopt changes in the environment. Complicated
trajectories are segmented into a sequence of attractors. A break point is inserted if LDS fails to predict
the next point. However it does not differentiate the subgoals which are and which are not associated
with the objects in the environment. \cite{LDMofTrajectories}
Sensory-motor learning
Reactive tasks learning using RBN: Passing through door passage and wall folowing using laser range
finder, \cite{behaviourLFD}
Reinforcement learning along with demonstration phase was applied to learn mapping from current
state to the next state and action \cite{RL2001}. The approach was tested for two simple tasks of wall-
following and obstacle avoidance.
Reinforcement with fuzzy \cite{ wallFolowing2009} for wall following.
Non-Position Target
Gaskett et al. \cite{visionTarget} introduced a RL-based approach for training a mobile robot to wander (obstacle avoidance) and
pursue a target using real-time vision. Docking task on Peoplebot was learned using RL \cite{ DockingTask2005}.
@article{wallFolowing2009, title={Reinforcement ant optimized fuzzy controller for mobile-robot wall- following control}, author={C. F. Juang and C. H. Hsu}, journal={ IEEE Transactions on Industrial Electronics}, volume={56(10)}, number={10}, pages={3931 - 3940}, year={2009}, publisher={IEEE} }
@inproceedings{bazierCurves, title={Mobile robots at your fingertip: Bezier curve on-line trajectory generation for supervisory control}, author={ J. H. Hwang and R. C. Arkin and D. S. Kwon}, booktitle={Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS’03}, volume={2}, pages={1444 - 1449}, year={2003}, organization={IEEE} }
@inproceedings{ijspeert2001, title={Trajectory formation for imitation with nonlinear dynamical systems}, author={ A.J. Ijspeert and J. Nakanishi and S. Schaal}, booktitle={Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS’01}, volume={2}, pages={752 - 757}, year={2001}, organization={IEEE} }
@inproceedings{LDMofTrajectories, title={Trajectory representation using sequenced linear dynamical systems}, author={K. R. Dixon and P. K. Khosla}, booktitle={ Proceedings of IEEE International Conference on Robotics and Automation, ICRA'04}, volume={4}, pages={3925 - 3930}, year={2004}, organization={IEEE} }
@inproceedings{trajectorywithSubgoals, title={Learning by observation with mobile robots: A computational approach},
author={K. R. Dixon and P. K. Khosla}, booktitle={Proceedings of IEEE International Conference on Robotics and Automation, ICRA'04}, volume={1}, pages={102 - 107}, year={2004}, organization={IEEE} }
@article{ behaviourLFD, title={A behavior-based mobile robot architecture for learning from demonstration}, author={M. Kasper and G. Fricke and K. Steuernagel and E. von Puttkamer}, journal={Robotics and Autonomous Systems}, volume={34(2)}, number={2}, pages={153 - 164}, year={2001}, publisher={Elsevier} }
@inproceedings{visionTarget, author={C. Gaskett, L. Fletcher, and A. Zelinsky}, title={Reinforcement learning for visual servoing of a mobile robot}, booktitle={Proceedings of Australian Conference on Robotics and Automation (ACRA2000), 2000}.
Year={2000}
@article{pegPushingandObstacleAvoidance, title={A reinforcement learning with evolutionary state recruitment strategy for autonomous mobile robots control}, author={Kondo, T. and Ito, K.}, journal={Robotics and Autonomous Systems}, volume={46}, number={2}, pages={111--124}, year={2004}, publisher={Elsevier} }
@inproceedings{RL2001, title={Effective reinforcement learning for mobile robots}, author={ W.D. Smart and L. Pack Kaelbling}, booktitle={Proceedings of IEEE International Conference on Robotics and Automation, ICRA'02}, volume={4}, pages={3404 - 3410}, year={2002},
