Practical Data-Induced Learning Control for Robotic Manipulators with Guaranteed Performance

This paper introduces a Practical Data-Induced Learning Control (P-DiLC) framework for high-precision motion control of multi-joint robotic manipulators. The proposed controller autonomously learns from operational data to handle challenges such as unknown inertial parameters (e.g., variable payloads), complex disturbances like friction, and external contact forces. A dual-timescale architecture enables the controller to build a predictive feedforward model for repeatable dynamics, such as gravitational torques, while adaptively suppressing unpredictable events. By systematically integrating a command filter, the design rigorously solves the fundamental problem of applying non-differentiable, data-induced learning laws to MIMO robotic systems. A composite energy function analysis proves that all signals are bounded and the tracking error converges to a user-prescribed neighborhood.