Agile But Safe: Learning Collision-Free High-Speed Legged Locomotion (2401.17583v3)

Abstract: Legged robots navigating cluttered environments must be jointly agile for efficient task execution and safe to avoid collisions with obstacles or humans. Existing studies either develop conservative controllers (< 1.0 m/s) to ensure safety, or focus on agility without considering potentially fatal collisions. This paper introduces Agile But Safe (ABS), a learning-based control framework that enables agile and collision-free locomotion for quadrupedal robots. ABS involves an agile policy to execute agile motor skills amidst obstacles and a recovery policy to prevent failures, collaboratively achieving high-speed and collision-free navigation. The policy switch in ABS is governed by a learned control-theoretic reach-avoid value network, which also guides the recovery policy as an objective function, thereby safeguarding the robot in a closed loop. The training process involves the learning of the agile policy, the reach-avoid value network, the recovery policy, and an exteroception representation network, all in simulation. These trained modules can be directly deployed in the real world with onboard sensing and computation, leading to high-speed and collision-free navigation in confined indoor and outdoor spaces with both static and dynamic obstacles.

• The paper introduces a dual-policy framework that combines agile and recovery modes to enable collision-free high-speed legged locomotion.
• It employs a reach-avoid value network based on Hamilton-Jacobi theory, integrating control-theoretic risk estimation with model-free RL.
• The ABS framework demonstrates robust performance in simulations and real-world tests, achieving speeds exceeding 3 m/s amid obstacles.

Overview of Agile But Safe Framework

The development of legged robots capable of agile and safe locomotion in cluttered environments is a significant step forward in robotics. A team at Carnegie Mellon University and ETH Zurich has developed a framework—dubbed Agile But Safe (ABS)—which empowers quadrupedal robots to exhibit high-speed locomotion while avoiding collisions in dynamic, cluttered spaces.

Agile Policy and Recovery Mechanism

ABS consists of two interplaying policies: an agile policy and a recovery policy. The agile policy is designed to execute rapid motor skills, enabling the robot to navigate at peak speeds exceeding 3 m/s amidst obstacles. However, agility alone cannot guarantee safety. To address potential failures, a recovery policy is implemented to intervene when collision risk is identified. This is a novel approach that combines the adaptability of model-free reinforcement learning (RL) with control-theoretic tools to achieve a balance between speed of operation and safety protocols.

Control-Theoretic Reach-Avoid Values

The integration of a control-theoretic component distinguishes ABS. The system employs a reach-avoid (RA) value network to estimate the risk associated with the agile policy. These RA values are learned based on Hamilton-Jacobi reachability theory and updated in real-time to guide the recovery policy. It results in a dual-policy setup where the RA values dictate whether the robot should continue with its primary agile behaviors or switch to the recovery policy to mitigate collision risks. Efficacious switching between these policies is crucial for maintaining high-speed locomotion while ensuring safety is not compromised.

Simulation and Real-World Deployment

The ABS framework has been rigorously trained and tested through simulation using terrain and obstacle diversity, and then deployed directly into real-world environments with minimal adaptation. Significantly, the system has demonstrated the ability to adapt to both static and dynamic obstacles, and to perform robustly in various real-world conditions, as supported by video evidence provided by the authors.

Conclusion

This research presents a fundamental advancement in legged robot navigation. By pioneering a safe, dynamic, and effective approach, the ABS framework ensures that high-speed quadrupedal locomotion does not lead to a compromise in safety. The impressive capabilities demonstrated by ABS, including navigation at an average speed of 2.1 m/s and a peak of 3.1 m/s through congested areas, mark an exciting momentum towards reliable autonomous legged systems that can potentially transcend structured laboratory conditions to operate effectively in the real world.