Continuous Bipedal Jumping via Sliding-Mode Regularized Predictive Control

Continuous bipedal jumping primarily involves consistently generating explosive movements and effectively managing landing impacts, which presents significant challenges. First, we propose a tracking controller based on sliding-mode regularized predictive control (SMRPC) to enable the robot stably return to its initial state along periodic trajectories and swiftly execute the next jump. For maintaining stable centroidal motions, the SMRPC generates a lumped wrench for the whole robot based on periodic trajectories. Then, the method distributes the lumped wrench to both foot surfaces by using regularized input. Second, we present a task-space compliance control to absorb impacts upon landing and manage transitions between ground and in-air phases. This task-level controller utilizes foot wrench feedback to incorporate vertical velocity compensation, thereby achieving whole-body compliance. Finally, the effectiveness of our control framework is demonstrated on a humanoid robot named BHR-B2, which successfully performs stable continuous jumps in both simulated and experimental environments.