Learning to Plan and Build Experience via Imitation in a Social Environment

N. Cuperlier, P. Laroque, P. Andry, and P. Gaussier (France)


Neural networks; Multi-agent systems; Cognitive processes; Imitation; Social interaction;


This paper focuses on how perception of others can be used as a basis for following behavior, and on the implication of this imitation on the global behavior of simulated software agents or real mobile robots. We report on how this simple mechanism, involving a strong link between learning and interaction, can be applied the same way from very sim ple to more complex architectures to obtain new behaviors via very simple social interactions. Our simulation results show the interaction impact on an autonomous agent popu lation performing navigation tasks. We insist on the global performance improvement of agents with such a follow ing behavior on a given task, using several efficiency evi dences which can be observed with higher level of imita tion behaviors, like search space reduction or enhancement of survivors state. On the other hand, we focus on the min imal set of skills required, in terms of cognitive and inter action capabilities, to enable the development of a common knowledge over a population of agents with this very lim ited social interaction.

Important Links:

Go Back