Terrain perception for a reconfigurable biomimetic robot using monocular vision

Abstract

This paper presents a reconfigurable biomimetic robot which is able to crawl and roll. The robot mimics the morphology of a huntsman spider that can transform between crawling and rolling by reconfiguring its legs. Terrain perception for reconfigurable biomimetic robots has not been studied in literature. This work tends to perceive and segment the terrain when the robot is crawling or in the steady state between rollings. A remote control system is designed with a server-client mechanism which can perform real-time image processing with GPGPU coding and develop a probabilistic framework for terrain perception. For validation, we test the system in both an indoor lab environment and a more uncontrolled outdoor environment. The results suggest that the system provides a trustable performance.