SAC-Loco: Safe and Adjustable Compliant Quadrupedal Locomotion

2025-09-27

AI Summary

Existing quadrupedal robot control methods lack adaptive, tunable compliance against external disturbances, leading to instability or falls under strong perturbations. This paper proposes a force-sensor-free online compliance modulation framework that integrates tunable-compliance reinforcement learning (RL), a Capture Point (CoP)-based safety policy, and a recoverability prediction network to enable dynamic switching between compliant locomotion and fall recovery. Innovatively, we introduce a teacher-student RL architecture and a neural-network-driven proactive policy-switching mechanism triggered prior to failure. Evaluated in simulation and on physical hardware, the method significantly enhances dynamic stability and motion safety under severe disturbances, enabling smooth recovery and continuous locomotion. To our knowledge, this is the first approach achieving real-time, physics-based-force-feedback-free compliance adaptation and robust policy transfer across perturbation regimes.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Natural Language Processing: Safety and Robustness
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Quadruped robots are designed to achieve agile locomotion by mimicking legged animals. However, existing control methods for quadrupeds often lack one of the key capabilities observed in animals: adaptive and adjustable compliance in response to external disturbances. Most locomotion controllers do not provide tunable compliance and tend to fail under large perturbations. In this work, we propose a switched policy framework for compliant and safe quadruped locomotion. First, we train a force compliant policy with adjustable compliance levels using a teacher student reinforcement learning framework, eliminating the need for explicit force sensing. Next, we develop a safe policy based on the capture point concept to stabilize the robot when the compliant policy fails. Finally, we introduce a recoverability network that predicts the likelihood of failure and switches between the compliant and safe policies. Together, this framework enables quadruped robots to achieve both force compliance and robust safety when subjected to severe external disturbances.

Problem

• Achieving adaptive compliance in quadruped robots under disturbances
• Providing tunable compliance without explicit force sensing
• Ensuring safe locomotion recovery from large external perturbations

Innovation

• Switched policy framework for compliant locomotion
• Force compliant policy without explicit force sensing
• Recoverability network predicting failure for policy switching