SATA: Safe and Adaptive Torque-Based Locomotion Policies Inspired by Animal Learning

2025-02-18

AI Summary

Legged robots exhibit poor compliance, low safety, and difficulty in sim-to-real transfer when deployed in human-robot coexisting environments due to conventional position-based control. Method: This paper proposes a torque-space adaptive locomotion strategy inspired by electromyographic signals and animal biomechanics. We introduce a novel framework integrating biomimetic modeling, adaptive curriculum learning, robust policy optimization, and torque-space end-to-end reinforcement learning—enabling, for the first time, biological-mechanism-guided online adaptive torque control. Contribution/Results: Our approach overcomes key limitations of prior methods—namely, inefficient exploration in nonlinear state spaces and weak cross-domain generalization—achieving zero-shot sim-to-real transfer. Experiments demonstrate significant improvements in compliance and safety on soft/slippery terrain, narrow passages, and under strong external disturbances, establishing a scalable paradigm for safe autonomous mobility in unknown, unstructured environments.

Technology Category

• Intelligent Robots: Behavior Learning & Control
• Natural Language Processing: Safety and Robustness
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions

Abstract

Despite recent advances in learning-based controllers for legged robots, deployments in human-centric environments remain limited by safety concerns. Most of these approaches use position-based control, where policies output target joint angles that must be processed by a low-level controller (e.g., PD or impedance controllers) to compute joint torques. Although impressive results have been achieved in controlled real-world scenarios, these methods often struggle with compliance and adaptability when encountering environments or disturbances unseen during training, potentially resulting in extreme or unsafe behaviors. Inspired by how animals achieve smooth and adaptive movements by controlling muscle extension and contraction, torque-based policies offer a promising alternative by enabling precise and direct control of the actuators in torque space. In principle, this approach facilitates more effective interactions with the environment, resulting in safer and more adaptable behaviors. However, challenges such as a highly nonlinear state space and inefficient exploration during training have hindered their broader adoption. To address these limitations, we propose SATA, a bio-inspired framework that mimics key biomechanical principles and adaptive learning mechanisms observed in animal locomotion. Our approach effectively addresses the inherent challenges of learning torque-based policies by significantly improving early-stage exploration, leading to high-performance final policies. Remarkably, our method achieves zero-shot sim-to-real transfer. Our experimental results indicate that SATA demonstrates remarkable compliance and safety, even in challenging environments such as soft/slippery terrain or narrow passages, and under significant external disturbances, highlighting its potential for practical deployments in human-centric and safety-critical scenarios.

Problem

• Improve safety in legged robot locomotion
• Enhance adaptability in unseen environments
• Achieve zero-shot sim-to-real transfer

Innovation

• Torque-based locomotion policies
• Bio-inspired adaptive learning
• Zero-shot sim-to-real transfer