State estimator for legged robots under diverse contact conditions

Abstract

This thesis proposes a state estimator for the legged robots under diverse contact conditions with the measurement of an inertial measurement unit (IMU) and leg kinematics while assuming static foot contact. The leg kinematics measurement is improved by using legged robot-specific methods. The Rejection of the Dynamic Contact method rejects the leg kinematics measurement when the estimated foot velocity exceeds a certain threshold. This method is further expanded to Covariance Adjustment for the Dynamic Contact method, which switches the uncertainty of the static contact assumption based on the estimated foot velocity, where the foot velocity within the time window of the smoother is also re-evaluated to enhance the effectiveness of the method. On the other hand, the Contact Loop method applies the static contact assumption as a temporal displacement cost when a foot contact is detected to be continuously stationary. Furthermore, the thesis proposes various types of state definitions (SO(3) and SE(k) groups) considering Lie group theory. Also, the group-affine property is exploited on the smoother-based legged robot state estimator. Because the proposed smoother is formulated with the residual functions with group-affine property, their Jacobians become independent from current state estimates. These state-independent Jacobians lead to better convergence properties in optimizing the cost in the smoother, especially under dynamic contact events. The proposed estimators are tested in various robots, including the Mini Cheetah robot, the Mechanum Wheel Legged Robot, and the Hound robot. The verification includes a simulation environment and real robot experiments, including online experiments, where the results are compared against state-of-the-art algorithms.