Safe Humanoid Autonomy and Teleoperation Framework
This paper presents the Safe Protective and Assistive Robot Kit (SPARK), a comprehensive toolbox designed to ensure safety in humanoid autonomy and teleoperation. Humanoid robots have gained significant popularity due to their versatility in interacting with complex environments. It is critical to ensure that no harm is caused during those interactions since unsafe behaviors can pose serious risks to the environment and human safety, considering humanoids' physical capabilities. Therefore, robust safety measures are essential in general humanoid robotics research and deployment. The inherently complex physical structures of humanoid robots further complicate the design of task-specific safety solutions. To alleviate this challenge, we introduce SPARK, a modular toolbox that integrates state-of-the-art safe control algorithms such as Safe Set Algorithm (SSA) into a generic robot control framework. SPARK enables users to configure safety behaviors across multiple dimensions, such as defining safety criteria and sensitivity levels, allowing for optimized trade-offs between safety and performance. The toolbox is natively compatible with the Unitree G1 humanoid robot, utilizing Apple Vision Pro (AVP) for perception, while also offering interfaces for seamless customization with alternative hardware setups. By integrating SPARK as a fail-safe mechanism, users can significantly improve the safety of their existing humanoid systems.
Kai S. Yun, Rui Chen, Chase Dunaway, John M. Dolan, Changliu
Liu
Submitted to ICRA 2025
@misc{sun2025spark,
title = {SPARK: A Toolbox for Safe Humanoid Autonomy and Teleoperation},
author = {Sun, Yifan and Chen, Rui and Yun, Kai S. and Fang, Yikuan and Jung, Sebin and Zhao, Weiye and Liu, Changliu},
year = {2025},
howpublished = {\url{https://intelligent-control-lab.github.io/spark/}}
}