Contact force estimation for a single leg test setup with compliance in MuJoCo

Abstract

Dynamic contact computation and force feedback becomes significantly challenging, when 5DoF single robot leg with compliant element follows fast dynamic trajectory. The objective of this work is to tune the MuJoCo contact parameters, based on experimental foot contact forces. Based on the tuned parameter, contact forces for squats and swing leg motion is simulated for foot-terrain contact and it is validated with experimentation for rigid, deformable and inclined terrains. Computer vision technique is utilized to record the terrain deformation to validate the MuJoCo contact model against the experimental results. Comparison of rigid and compliant shank during contact in a testbed setup is studied. Robot foot shoe along with electronics for measuring normal contact force is designed. The robot dynamics have been modeled using MuJoCo dynamics model and PD control with feedforward gain is implemented, to ensure single robot leg follows the desired trajectory. Moreover, the deformation model is utilized to model the compliant shank, based on its optimal structural parameters. The robot is simulated using MuJoCo physics simulation, which requires robot’s xml in mjcb format to follow joint trajectories. Simulation contact forces are compared with experimental results for rigid and compliant shank. The experimental results confirm that there is 11.67% reduction in contact forces and better trajectory following due to addition of 4 mm compliant bracket in robot leg tested in rigid wooden terrain. Robot squat on PU foam records 12.86% reduction in contact force and 26 mm terrain deformation. The results demonstrate that the proposed leg design outperforms its rigid counterpart.