A generic optimization-based framework for reactive collision avoidance in bipedal locomotion

In this work we present a novel and generic framework for reactive collision avoidance in bipedal locomotion, which is formulated as an optimization problem considering the constraints of collision avoidance as well as others (e.g. joint limits) to simultaneously satisfy both Cartesian and joint space objectives. To realize the reactive behaviors, several task space motions, such as the translational motion of the swing foot and the vertical position of the support foot, could be relaxed in presence of obstacles. Therefore, the swing foot trajectory is modulated with respect to the references in real-time for preventing future collisions between the legs, or legs and obstacles in the environment. External obstacle negotiation in the proposed framework can also be addressed generically by treating the obstacle as an extended segment of the support foot. The allowable deviation of the relaxed degrees of freedom from their references could be further utilized to modify the foot placement to regenerate a reactive walking pattern. The validation and the performance of the proposed method are fully evaluated and demonstrated in physics based simulations of the compliant humanoid robot COMAN.