Neural Internal Model Control: Learning a Robust Control Policy via Predictive Error Feedback

TL;DR

Robust robotic motion control remains challenging under nonlinear disturbances. The authors present NeuralIMC, a framework that marries Internal Model Control with a model-free RL policy by using a simplified rigid-body predictive model to generate predictive error signals for feedback. The approach is demonstrated on quadrotors and quadrupedal robots, showing superior robustness and generalization, including successful sim-to-real transfer with rope-suspended payloads. The work provides interpretable, plug-and-play robustness gains and releases its code for broader adoption.

Abstract

Accurate motion control in the face of disturbances within complex environments remains a major challenge in robotics. Classical model-based approaches often struggle with nonlinearities and unstructured disturbances, while RL-based methods can be fragile when encountering unseen scenarios. In this paper, we propose a novel framework, Neural Internal Model Control, which integrates model-based control with RL-based control to enhance robustness. Our framework streamlines the predictive model by applying Newton-Euler equations for rigid-body dynamics, eliminating the need to capture complex high-dimensional nonlinearities. This internal model combines model-free RL algorithms with predictive error feedback. Such a design enables a closed-loop control structure to enhance the robustness and generalizability of the control system. We demonstrate the effectiveness of our framework on both quadrotors and quadrupedal robots, achieving superior performance compared to state-of-the-art methods. Furthermore, real-world deployment on a quadrotor with rope-suspended payloads highlights the framework's robustness in sim-to-real transfer. Our code is released at https://github.com/thu-uav/NeuralIMC.

Figures (5)

• Figure 1: Predictive errors can be caused by mismatched system modeling and/or external disturbances.
• Figure 2: The closed-loop NeuralIMC framework using an explicit predictive model to calculate the next body state, applirope to both quadrotors and legged robots.
• Figure 3: Feedback structures of classical feedback control and Internal Model Control (IMC).
• Figure 4: Ablation study on the feedback structure with different history lengths.
• Figure 5: Real-world experiments on a quadrotor with two different rope-suspending payloads.