Reference-Augmented Learning for Precise Tracking Policy of Tendon-Driven Continuum Robots

2026-04-28

AI Summary

This work addresses the challenging trajectory tracking problem in tendon-driven continuum robots (TDCRs), which arises from their highly nonlinear, path-dependent, and non-Markovian dynamics. To overcome this, the authors propose a reference-augmented offline learning framework that employs a differentiable recurrent neural network (RNN) as a surrogate dynamics model to serve as a gradient bridge for optimization. The approach integrates a multi-scale reference augmentation strategy—comprising random offsets, harmonic perturbations, and random walks—to internalize diverse error-recovery capabilities without requiring additional online interaction. Experimental validation on a three-segment TDCR platform demonstrates that the proposed method reduces average position error by 50.9% compared to baseline approaches and substantially outperforms conventional Jacobian-based controllers. Notably, it exhibits strong generalization and robustness on out-of-distribution trajectories, achieving high-precision end-to-end six-degree-of-freedom tracking control.

Abstract

Tendon-Driven Continuum Robots (TDCRs) pose significant control challenges due to their highly nonlinear, path-dependent dynamics and non-Markovian characteristics. Traditional Jacobian-based controllers often struggle with hysteresis-induced oscillations, while conventional learning-based approaches suffer from poor generalization to out-of-distribution trajectories. This paper proposes a reference-augmented offline learning framework for precise 6-DOF tracking control of TDCRs. By leveraging a differentiable RNN-based dynamics surrogate as a gradient bridge, we optimize a control policy through an augmented reference distribution. This multi-scale augmentation scheme incorporates stochastic bias, harmonic perturbations, and random walks, forcing the policy to internalize diverse tracking error recovery mechanisms without additional hardware interaction. Experimental results on a three-section TDCR platform demonstrate that the proposed policy achieves a 50.9\% reduction in average position error compared to non-augmented baselines and significantly outperforms Jacobian-based methods in both precision and stability across various speeds.

Problem

• Tendon-Driven Continuum Robots
• nonlinear dynamics
• hysteresis
• out-of-distribution generalization
• tracking control

Innovation

• reference-augmented learning
• tendon-driven continuum robots
• differentiable RNN surrogate
• offline policy optimization
• multi-scale trajectory augmentation