A sensorless torque control for Antagonistic Driven Compliant Joints

Antagonistic Driven Compliant Joints (ADCJs) are object of great interest in current robotics research, representing one of the most widely applied solutions to develop human-like and safe joints for human-robot interaction. Providing the joint with “actively” adjustable hardware compliance, ADCJs have two distinctive features: (1) the joint is powered by two independent “actuation units” and (2) each actuation unit works as a non-linear elastic element with an adjustable resting position. This paper proposes a sensorless torque control strategy suitable for ADCJs actuated robots. This method is based on two steps: (1) off-line characterization of the elasticity of the actuation units, defined by the force–elongation curve and (2) online estimation of the force exerted by each actuation unit, through a direct measure of the joint angle, and of the “resting position” of each actuation unit. The proposed force estimation method can be used to develop two independent force controllers, which can be then combined to regulate the resulting joint torque, with no need of additional torque sensors. The performance of the proposed torque control was evaluated over the shoulder and the elbow ADCJs of the 2-link 2-DOFs planar robotic arm NEURARM. The method proved to work effectively, achieving good performances on the test platform, and represents a suitable alternative to state-of-the-art sensor-based torque controls.