A Mapping Approach to Achieve Torque Control for Parallel-Actuated Robotic Systems

In this work, we present an approach for realizing the torque control for a parallel-actuated robotic system by mapping the motion of a linear series elastic actuator (LSEA) to its driven robot joint. In most standard robotic modeling and control strategies, a robot is assumed to be actuated by torques applied directly at each joint and constructed as an open kinematic chain. However, the use of non-direct-drive actuators can violate these assumptions, causing additional challenges for the modelling and control of the robot. On our humanoid robot we use standard high level controllers to command desired joint positions and torques in order to generate desired behaviors. However, the humanoid robot is actually actuated by LSEAs, which are defined by actuator length and force. Overcoming this difference requires a method of mapping the motion and effort of an LSEA onto the corresponding joint of a robot. Our mapping approach allows for the conversion of generic desired joint position and torque trajectories consistent with standard controllers into actuator length and force trajectories that can be implemented on an LSEA-actuated robot. We present a two-stage methodology to achieve low-level torque control on our humanoid robot: a validation of the force-torque mapping in simulation, and a force controller implementation for tracking these resulting torque trajectories on a sample simulation of a single robot joint.