Responsive navigation of a biped robot that takes into account terrain, foot-reachability and capturability

Abstract

A control architecture for robust biped navigation that takes into account traversability on terrain, foot-reachability, and capturability all at once in dynamic situations is proposed. It consists of online mapping of terrain and quick decision of foot placement. A grid map with elevation, normal direction and the standard deviation of measurement is proposed for the former and is named the stochastic Elevation-and-Normal Grid (ENG) map. It is incrementally updated as depth sensor information comes in every cycle. The traversability of each grid can be easily checked based on the associated information. The latter stands upon the canonical biped track (CBT), which is a pair of continuous paths that the feet track. It is drawn along a nominal collision-free path on the ENG map. The most efficient foot placement at the next step from the viewpoint of the foot-reachability is chosen on them as the initial candidate, around which a feasible placement accommodated with traversability and capturability is found by the breadth-first search. An autonomous biped locomotion system is completed by combining the above architecture with a foot-guided controller based on the terminal capturability. The efficacy of the proposed method was validated through some computer simulations.