1A1-F14 Robot control inspired by Escherichia Coli chemotaxis

概要

論文の詳細を見る
Living beings like bacteria search food using extremely simple strategies that reveal to be very robust. From this observation we derive an algorithm for robot control. Most literature on the topic defines basic robot behaviors that are used to mimic bacteria movement in the physical space. This paper proposes instead to work directly in the motor command space, allowing the robot to determine itself the control signals to use. The strategy underlying the algorithm consists in simply repeating control signals that lead to condition improvements. Counterintuitively, adding random perturbations to the control signal being repeated improves the performances. A model of the phenomenon is provided and an algorithm for automatic choosing the perturbation magnitude is proposed. Results of experiments on the robustness and practical applicability of the approach are reported. In particular we show that the algorithm is able to perform robot navigation even when no information on the robot structure is available and the robot undergoes hardware damages.
一般社団法人日本機械学会の論文