Online Multi-Phase Trajectory Generation for Compliant Landing Control of Quadruped Robots

While modern quadruped robots can jump or even somersault, without proper control, landing from the air could cause serious impacts and damages to both the mechanical and electrical components of robots. The landing process often faces different robot's poses and velocities, e.g., vertical, horizontal, and/or pitch movements, as well as contact phase change, e.g., from initial two-leg contact to final four-leg contact phase. Also, the time left for the robot to respond to emerging contacts is extremely short before giving rise to any damage. Considering all these challenges, this paper proposes a compliant landing control framework for quadruped robots. First, based on a 2-D rigid body model, we formulate the computing of reference landing trajectories with as much compliance as possible as a quadratic programming problem, which can be solved quickly online. Then, a model-predictive control algorithm is used to determine the joint torque commands for following the reference trajectories to realize compliant landing. The effectiveness of the proposed framework has been verified with various landing scenarios in both simulation and real experiments.