Dual-Arm Box Grabbing With Impact-Aware MPC Utilizing Soft Deformable End-Effector Pads
Niels Dehio, Yuquan Wang, Abderrahmane Kheddar
Abstract
Safely generating impacts is challenging due to subsequent discontinuous velocity jumps and high impact forces. For this reason, state-of-the-art multi-robot controllers performing collaborative object manipulation typically approach contacts with low relative velocities. We instead aim for high-speed bi-manual impact-aware swift grabbing tasks, generating intentionally maximum feasible impacts. Therefore, we propose that the robot end-effector tips be equipped with deformable soft pads. The hardware modification partially absorbs the shock and enables us to control the deformation state during the prolonged impact duration. Mapping the robot's structural hardware constraints from the high-dimensional configuration-space to the low-dimensional contact-space enables us to model the bounded deformation dynamics explicitly. Exploiting constrained model-predictive control, we maximize the impact velocity within the feasible limits for dual-arm setups with deformable end-effector tips. Our control paradigm is assessed with real-robot experiments on two redundant Panda manipulators, demonstrating high pre-impact velocities for boxes with varying weights and size. The approach also applies to grabbing soft objects with rigid end-effectors.