Predictive Control with Indirect Adaptive Laws for Payload Transportation by Quadrupedal Robots
This study addresses the challenge of stable payload transportation for quadrupedal robots under unknown or dynamic loads, model uncertainties, and complex terrains. The authors propose a hierarchical planning and control framework: at the high level, a gradient-based indirect adaptive law is integrated with model predictive control (MPC) to online estimate parameters of a reduced-order motion model and generate real-time trajectories; at the low level, a nonlinear whole-body controller tracks these trajectories. The approach innovatively combines indirect adaptation with MPC and incorporates convex stability constraints to ensure convergence of parameter estimation errors. Experimental results demonstrate that the system can transport static unknown payloads up to 109% and 91% of its body weight on flat and rough terrain, respectively, as well as dynamic payloads up to 73% of its weight. Hardware tests confirm robustness against disturbances, obstacles, and outdoor conditions, significantly outperforming conventional MPC and L1-MPC baselines.