Leg State Estimation for Quadruped Robot by Using Probabilistic Model With Proprioceptive Feedback

Legged robots are sent into outdoor environments and desired to explore unstructured terrains like animals in nature. Therefore, the ability to robustly detect leg phase transitions should be a critical skill. However, many current approaches rely on external sensors mounted on legged robots, which increases the overall cost or renders the robot useless if the sensors fail. Conversely, when a robot's proprioceptive sensors fail, its ability to control its motion is compromised. Therefore, as long as the robot is capable of locomotion, the proprioceptor-based leg state estimation method can be applicable. Based on this feature, we propose a novel leg phase detection method for quadruped robots that uses proprioceptive feedback to estimate leg state while overcome the problem of inaccurate in the absence of external devices. The innovative estimation method deftly identifies leg phases even in the absence of a priori terrain features, allowing the robot to traverse the terrain without prior knowledge or reliance on vision-based detection. Through extensive hardware experiments in different scenarios, the proposed approach demonstrates robust estimation of leg states.