Multistrategy Learning Methods for Multirobot Systems

Abstract : Incorporation of a range of disparate learning algorithms is both feasible and desirable within a hybrid deliberative/reactive architecture. In particular, the authors present three different methods suitable for multi-robot missions: learning momentum, a parametric adjustment technique; Q-learning of roles when represented as behavioral assemblages in the context of team performance; and a case-based wizard to enhance the user's ability to specify complex multi-robot missions. Future work involves expanding other learning algorithms already in use for single robot missions including as well as investigating the interactions between these methods when they are allowed to be active concurrently. A range of simulation experiments and results are reported using the Georgia Tech MissionLab mission specification system.

[1]  Ronald C. Arkin,et al.  Multiagent Mission Specification and Execution , 1997, Auton. Robots.

[2]  Ronald C. Arkin,et al.  Usability evaluation of high-level user assistance for robot mission specification , 2004, IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews).

[3]  Maja J. Mataric,et al.  Reward Functions for Accelerated Learning , 1994, ICML.

[4]  Ronald C. Arkin,et al.  Learning momentum: integration and experimentation , 2001, Proceedings 2001 ICRA. IEEE International Conference on Robotics and Automation (Cat. No.01CH37164).

[5]  Ronald C. Arkin,et al.  Robot behavioral selection using q-learning , 2002, IEEE/RSJ International Conference on Intelligent Robots and Systems.

[6]  Tucker R. Balch,et al.  Communication in reactive multiagent robotic systems , 1995, Auton. Robots.

[7]  Chris Watkins,et al.  Learning from delayed rewards , 1989 .

[8]  Richard S. Sutton,et al.  Reinforcement Learning: An Introduction , 1998, IEEE Trans. Neural Networks.