Presentation Information
[C-10_C-11-05]Improvement in Ground Condition Perception of Amoeba-inspired Autonomous Walking Robot
〇Hyoto Yamaguchi1, Zenji Yatabe1, Seiya Kasai1 (1. Hokkaido University)
Keywords:
Autonomous walking robot,Amoeba-inspired solution search,Four-legged robot,Reservoir computing
There is a growing demand for autonomous robots capable of operating in harsh environments with poor visual information, such as disaster sites. We have been developing an amoeba-inspired autonomous walking robot that walks autonomously by successively searching for and executing appropriate leg movements in response to the situation on site, without being taught how to walk in advance. At present, because the robot searches for leg movement at every step, its behavior is not organized as a walking pattern, and the locomotion efficiency is low. Therefore, we aim to realize the ability of the robot itself to identify the ground condition and acquire efficient walking gait patterns depending on the ground condition. As a means of ground condition perception, we employ artificial proprioception without relying on vision, and investigate a method for perceiving the ground condition by analyzing sensor time-series data using reservoir computing (RC). In this study, to improve the classification ability, we evaluated and analyzed the multimodal sensor time-series data and the contribution of each sensor to ground condition perception.