人型2脚歩行ロボットの斜面歩行に関する研究

ステレオビジョンによる斜度推定とその斜面歩行への応用

抄録

The bipedal walk is a useful method for moving in human live environment. However, there are few research works on biped robots on uneven ground though adaptability to those environments is necessary for them. In this paper, the slope-walk of a biped robot based on stereo vision is described. First, a method to estimate inclination with vision is proposed. This method is based on triangulation. The robot observes the ground in front of it at a certain angle below horizon and measures distance to there with stereo vision. This distance becomes smaller on uphill slope and larger on downhill way. The vision is also used for target detection for followed-walk. The robot detects distance and direction of a target. We propose direct utilization of parallax as distant information and direct conversion from parallax to inclination. These methods provide enough accuracy for walking robot and reduce cost of calculations and calibrations. Secondly, a sensor based walk is proposed. The sequence of walk consists of two parts. One is stabilization phase that absorbs attitude-disturbances of robot caused by environment or alteration of gait. This stabilization uses attitude sensors mounted on the body. The other is stepping phase. We designed simple gaits so that it can be altered easily for slope-walk and curve-walk. The experiments using a biped robot with legs of human type degree of freedoms and a stage that provides various slopes were carried out. The accuracy of slope estimation was about ±1[deg] and the walk through undulating ground of 5[deg] downhill to 5[deg] uphill way was achieved. The results showed that our method gains adaptability of biped robots to slopes.