Composite Learning Tracking Control of High-DoF Compliant-Joint Industrial Robots

Respecting joint compliance during the control design is significant for industrial robots to improve tracking accuracy and transient behavior. This study aims to achieve tracking control for industrial robots with high degrees of freedom (DoFs) and nonnegligible joint compliance. By introducing a novel parameterized representation, we proposed a singular perturbation (SP)-based composite learning robot control (CLRC) strategy for compliant-joint robots to achieve practical exponential stability, where parameter convergence is ensured without the strict condition of persistent excitation. An effective algorithm is provided to find the identifiable parameter set containing actuator inertia parameters, so it is possible to mix composite learning with SP-based adaptive control to make the latter feasible for compliant-joint robots. The proposed SP-CLRC does not depend on joint acceleration and jerk feedback and has a low implementation cost comparable to its rigid-joint counterpart. Experiments on a compliant-joint industrial robot with seven DoFs named Franka Emika Panda have shown that the proposed method has a strong learning capacity and greatly outperforms state-of-the-art controllers in trajectory tracking and model prediction after only short-time one-shot learning.