Behaviour Learning by a Reward-Penalty Algorithm : From Gait Learning to Obstacle Avoidance by Neural Networks
Designing a mobile robot capable to evolve on various floors is a hard, and not yet solved task. Among the numerous ways, the biological inspired researches are attractive. Taking inspiration from such devices, a study on walk learning and obstacle avoiding by neural networks is presented. After a general presentation of the learning method, and its adaptation to neural networks, results on gait learning are described and first simulations on obstacle avoidance are also presented. Finally, a global network performing the two tasks with success is given.
KeywordsMobile Robot Obstacle Avoidance Learn Automaton Parallel Iteration Legged Robot
Unable to display preview. Download preview PDF.
- 3.A. Frank, “Walkers”. “Encyclopaedia of robotics: application and automation” DGRF R.C. ed. Wiley Interscience, (1988).Google Scholar
- 4.Vehicles”. V. Braitenberg; MIT Press 1984.Google Scholar
- 6.C. Touzet, O. Sarzeaud. “Application d’un algorithme d’apprentissage á la génération de formes locomotrices hexapodes”. Journées de Rochebrune AFCET (1992).Google Scholar
- 8.A. Johannet, I. Sarda. “Gait Learning of an Hexapod Robot: from Simulation to Realization”. Proceedings of the 2nd Japan-France Congress on Mechatronics. Takamatsu, Japan,(1994).Google Scholar