Humanoid robots have been rapidly advancing in recent years. Their human-like structure enables them to assist with tasks in a manner similar to humans and even learn dexterous skills through human demonstration. However, the high dimensionality of humanoid systems presents significant challenges. At the Intelligent Control Lab, our mission is to develop general whole-body control strategies for humanoid robots while ensuring safety in their operation. Our ultimate goal is to enable humanoid robots to assist humans and autonomously perform tasks in both daily life and industrial settings in a safe and efficient manner.

Safety is a critical concern for robots, encompassing the protection of the robot itself, its interaction with the environment, and, most importantly, its interaction with humans. For humanoid robots, ensuring safety is especially challenging and requires careful consideration.

To support the safe deployment of complex robotic systems, SPARK serves as a modular and composable toolbox that integrates state-of-the-art safe control algorithms. Users can easily configure safety criteria and sensitivity levels to optimize the trade-off between safety and performance. To accelerate research and development in humanoid robot safety, SPARK provides a simulation benchmark that enables the comparison of safety approaches across various environments, tasks, and robot models. Additionally, it facilitates the rapid deployment of synthesized safe controllers on real robots. For hardware implementation, SPARK supports external sensors such as Apple Vision Pro (AVP) or a Motion Capture System, while also offering interfaces for seamless integration with alternative hardware setups. This paper demonstrates SPARK’s capabilities through both simulation experiments and real-world case studies using a Unitree G1 humanoid robot. By leveraging SPARK, users and researchers can significantly enhance the safety of humanoid systems while accelerating advancements in the field.

Publications:

1. [U] SPARK: A Modular Benchmark for Humanoid Robot Safety
   Yifan Sun, Rui Chen, Kai S Yun, Yikuan Fang, Sebin Jung, Feihan Li, Bowei Li, Weiye Zhao and Changliu Liu
   arXiv preprint arXiv:2502.03132, 2025
   

   Preprint Code Video Abstract Cite

   Citation Formats: BibTeX Plain Text

   Abstract:
   

   This paper introduces the Safe Protective and Assistive Robot Kit (SPARK), a comprehensive benchmark designed to ensure safety in humanoid autonomy and teleoperation. Humanoid robots pose significant safety risks due to their physical capabilities of interacting with complex environments. The physical structures of humanoid robots further add complexity to the design of general safety solutions. To facilitate the safe deployment of complex robot systems, SPARK can be used as a toolbox that comes with state-of-the-art safe control algorithms in a modular and composable robot control framework. Users can easily configure safety criteria and sensitivity levels to optimize the balance between safety and performance. To accelerate humanoid safety research and development, SPARK provides a simulation benchmark that compares safety approaches in a variety of environments, tasks, and robot models. Furthermore, SPARK allows quick deployment of synthesized safe controllers on real robots. For hardware deployment, SPARK supports Apple Vision Pro (AVP) or a Motion Capture System as external sensors, while also offering interfaces for seamless integration with alternative hardware setups. This paper demonstrates SPARK’s capability with both simulation experiments and case studies with a Unitree G1 humanoid robot. Leveraging these advantages of SPARK, users and researchers can significantly improve the safety of their humanoid systems as well as accelerate relevant research. The open-source code is available at (https://github.com/intelligent-control-lab/spark)

   Video:
   

In this paper, we address the problem of dexterous safety, which enforces limb-level geometric constraints to prevent both external and self-collisions in cluttered environments. Unlike traditional safety approaches that rely on simplified bounding geometries in sparse environments, dexterous safety introduces numerous constraints, often leading to infeasible constraint sets when solving for safe robot control. To tackle this challenge, we propose the Projected Safe Set Algorithm (p-SSA), an extension of classical safe control algorithms designed for multi-constraint scenarios. p-SSA systematically relaxes conflicting constraints in a principled manner, minimizing safety violations while ensuring feasible robot control. We validate our approach both in simulation and on a real Unitree G1 humanoid robot performing complex collision avoidance tasks. Experimental results demonstrate that p-SSA enables the humanoid to operate robustly in challenging environments with minimal safety violations, while generalizing across various tasks with zero parameter tuning.

Publications:

1. [U] Dexterous Safe Control for Humanoids in Cluttered Environments via Projected Safe Set Algorithm
   Rui Chen, Yifan Sun and Changliu Liu
   arXiv preprint arXiv:2502.02858, 2025
   

   Preprint Abstract Cite

   Citation Formats: BibTeX Plain Text

   Abstract:
   

   It is critical to ensure safety for humanoid robots in real-world applications without compromising performance. In this paper, we consider the problem of dexterous safety, featuring limb-level geometry constraints for avoiding both external and self-collisions in cluttered environments. Compared to safety with simplified bounding geometries in sprase environments, dexterous safety produces numerous constraints which often lead to infeasible constraint sets when solving for safe robot control. To address this issue, we propose Projected Safe Set Algorithm (p-SSA), an extension of classical safe control algorithms to multi-constraint cases. p-SSA relaxes conflicting constraints in a principled manner, minimizing safety violations to guarantee feasible robot control. We verify our approach in simulation and on a real Unitree G1 humanoid robot performing complex collision avoidance tasks. Results show that p-SSA enables the humanoid to operate robustly in challenging situations with minimal safety violations and directly generalizes to various tasks with zero parameter tuning.

Controlling legged robots, especially humanoids and quadrupeds, is challenging due to their high-dimensional, nonlinear dynamics. While Model Predictive Control (MPC) works well for linear systems, nonlinear control remains complex. The Koopman Operator offers a potential solution by approximating nonlinear dynamics with a linear model, but issues like approximation errors and domain shifts limit its scalability. This paper proposes a continual learning algorithm that iteratively refines Koopman dynamics by expanding the dataset and latent space, improving approximation accuracy. Theoretical analysis shows monotonic error convergence, and experiments on Unitree G1/H1/A1/Go2 and ANYmal D demonstrate robust locomotion control across diverse terrains using simple linear MPC. This work is the first to successfully apply linearized Koopman dynamics to high-dimensional legged robots, enabling scalable model-based control.

Publications:

1. [U] Continual Learning and Lifting of Koopman Dynamics for Linear Control of Legged Robots
   Feihan Li, Abulikemu Abuduweili, Yifan Sun, Rui Chen, Weiye Zhao and Changliu Liu
   arXiv:2411.14321, 2024
   

   Preprint Page Video Abstract Cite

   Citation Formats: BibTeX Plain Text

   Abstract:
   

   The control of legged robots, particularly humanoid and quadruped robots, presents significant challenges due to their high-dimensional and nonlinear dynamics. While linear systems can be effectively controlled using methods like Model Predictive Control (MPC), the control of nonlinear systems remains complex. One promising solution is the Koopman Operator, which approximates nonlinear dynamics with a linear model, enabling the use of proven linear control techniques. However, achieving accurate linearization through data-driven methods is difficult due to issues like approximation error, domain shifts, and the limitations of fixed linear state-space representations. These challenges restrict the scalability of Koopman-based approaches. This paper addresses these challenges by proposing a continual learning algorithm designed to iteratively refine Koopman dynamics for high-dimensional legged robots. The key idea is to progressively expand the dataset and latent space dimension, enabling the learned Koopman dynamics to converge towards accurate approximations of the true system dynamics. Theoretical analysis shows that the linear approximation error of our method converges monotonically. Experimental results demonstrate that our method achieves high control performance on robots like Unitree G1/H1/A1/Go2 and ANYmal D, across various terrains using simple linear MPC controllers. This work is the first to successfully apply linearized Koopman dynamics for locomotion control of high-dimensional legged robots, enabling a scalable model-based control solution.

   Video: