Fault-Tolerant Physical Human-Robot Interaction via Stiffness Adaptation of Elastic Actuators

Elastic actuators are popular in human-robot interaction as they can improve human safety and efficiency. Yet, such actuators are more complex than rigid ones and might be subject to additional technical faults, e.g., stiffness changes. This paper extends previous studies on stiffness-fault-tolerant physical human-robot interaction (pHRI) through control adaptation, introducing new methods for stiffness estimation and fault evaluation. Kalman filters with different measurement signals and system models estimating the actual stiffness value of the elastic element are compared. Faults are evaluated by analyzing the structural durability and compensated by adapting an impedance controller to provide a desired interaction stiffness. Experiments with a series elastic actuator underline the feasibility of the evaluation and compensation methods for attaining safe and reliable pHRI. Results show that stiffness estimation during pHRI is possible when the actuator friction and interaction torque is either negligible or well known, or when the torque at the spring is measured