Humanoid Robots: Safety and Dexterity

WHY HUMANOIDS?

Human-shaped robots need whole-system intelligence

Humanoid robots can navigate human spaces, use human tools, and learn from human motion—but their high-dimensional dynamics and rapidly changing contacts make every useful behavior a control and safety challenge.

Our research treats safety, control, learning, and manipulation as connected parts of one system. We study how a robot can model its dynamics, coordinate its body, recognize when recovery remains possible, transfer behaviors to hardware, and combine reusable skills into useful work.

Safety is both a controller and a prediction problem. SPARK and p-SSA enforce geometric constraints, while PRISM and λ-Reachability characterize states from which safe behavior remains possible.

Explore safety research

HUMANOID SAFETY

Protect the robot before failure becomes unavoidable

Safety spans fast control corrections, state-level monitors, and reusable infrastructure that carries safety methods from simulation to hardware.

OPEN SAFETY INFRASTRUCTURE

SPARK

Safe Protective and Assistive Robot Kit

SPARK is an open-source benchmark and toolbox for configuring safety criteria, comparing safe-control approaches in simulation, and deploying synthesized controllers on complex robot systems.

Its modular framework supports alternative sensing setups and is demonstrated through simulation studies and Unitree G1 hardware case studies.

Composable safety constraints
Simulation benchmarking
Real-hardware deployment

Project page Open-source code

SPARK demonstrations of human-aware control, collision avoidance, and balance recovery — Safety scenarios spanning human-aware interaction, collision avoidance, and whole-body recovery.

Publication

[C101] SPARK: Safe Protective and Assistive Robot Kit
Yifan Sun, Rui Chen, Kai S Yun, Yikuan Fang, Sebin Jung, Feihan Li, Bowei Li, Weiye Zhao and Changliu Liu
IFAC Symposium on Robotics, 2025
Citation Formats:
```
    
```
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:

Projected Safe Set Algorithm preventing arm collisions during humanoid teleoperation — p-SSA resolves competing geometric constraints while retaining a feasible control command.

DEXTEROUS SAFETY

Safe motion in cluttered environments

Dexterous safety can require many simultaneous limb-level constraints for external and self-collision avoidance. The Projected Safe Set Algorithm (p-SSA) relaxes conflicting constraints by minimizing safety violations while preserving command feasibility.

Publication

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

SAFETY BEFORE THE LAST RESORT

Safe-stoppability monitoring

PRISM

Learn where safe stopping remains possible

PRISM formalizes emergency stopping as a policy-dependent question and uses importance sampling to refine a neural stoppability monitor near rare, safety-critical boundary states.

Project page

Publication

[C115] Learning Safe-Stoppability Monitors for Humanoid Robots
Yifan Sun, Yiyuan Pan, Shangtao Li, Caiwu Ding, Tao Cui, Lingyun Wang and Changliu Liu
IEEE/RSJ International Conference on Intelligent Robots and Systems, 2026
Citation Formats:
```
    
```
Abstract:

Emergency stop (E-stop) mechanisms are the de facto standard for robot safety. However, for humanoid robots, abruptly cutting power can itself cause catastrophic failures; instead, an emergency stop must execute a predefined fallback controller that preserves balance and drives the robot toward a minimum-risk condition. This raises a critical question: from which states can a humanoid robot safely execute such a stop? In this work, we formalize emergency stopping for humanoids as a policy-dependent safe-stoppability problem and use data-driven approaches to characterize the safe-stoppable envelope. We introduce PRISM (Proactive Refinement of Importance-sampled Stoppability Monitor), a simulation-driven framework that learns a neural predictor for state-level stoppability. PRISM iteratively refines the decision boundary using importance sampling, enabling targeted exploration of rare but safety-critical states. This targeted exploration significantly improves data efficiency while reducing false-safe predictions under a fixed simulation budget. We further demonstrate sim-to-real transfer by deploying the pretrained monitor on a real humanoid platform. Results show that modeling safety as policy-dependent stoppability enables proactive safety monitoring and supports scalable certification of fail-safe behaviors for humanoid robots.

Video:

GEOMETRIC-HORIZON SAFETY LEARNING

λ-Reachability

Geometric-Horizon Safety Bellman Equations for Humanoid Safety

Lambda-Reachability hardware experiments predicting non-recoverable states before collision and balance failures — Hardware safety-value inference for dynamic-obstacle avoidance and stable locomotion. Orange arrows mark the predicted first non-recoverable state; red arrows mark the later unsafe event.

SCALABLE REACHABILITY ANALYSIS

Predict future safety violations

λ-Reachability learns a safety value that represents the worst future safety signal under a policy. A geometric rollout horizon lets the learning target interpolate between local one-step updates and longer-horizon safety information.

With terminal survival parameter δ<1, the resulting safety Bellman operator is a contraction. As λ approaches 1, its fixed point approaches the undiscounted reachability objective. The paper evaluates balance and collision-avoidance constraints on simulated and physical humanoids.

Read arXiv:2606.16022 Download PDF

Variable horizonMix short- and long-horizon max targets through an interpretable parameter.

Stable learningRetain a contraction mapping for δ<1.

Humanoid validationEvaluate balance and collision safety in simulation and on hardware.

Publication

[C118] λ-Reachability: Geometric-Horizon Safety Bellman Equations for Humanoid Safety
Rui Chen, Shangtao Li, Yifan Sun and Changliu Liu
Conference on Robot Learning, 2026
Citation Formats:
```
    
```
Abstract:

We introduce λ-Reachability, a scalable approach to Hamilton–Jacobi safety analysis for high-dimensional robotic systems. Unlike prior discounted formulations that rely on fixed one-step Bellman updates, λ-Reachability employs a stochastic multi-step estimator of the safety value, using a geometrically distributed rollout horizon together with a randomly absorbed terminal. Conceptually analogous to TD(λ), λ-Reachability interpolates between local self-consistency updates and long-horizon max-over-trajectory safety targets via an interpretable horizon-control parameter. Unlike TD(λ), where the terminal value is always incorporated in learning targets, the terminal safety value in λ-Reachability is only used at a probability controlled by parameter δ. We formally show that for δ<1, the update induces a contraction mapping that allows temporal-difference learning; as λ\to 1, the estimator recovers the undiscounted reachability objective. We apply λ-Reachability to high-dimensional safety learning problems with both simulated and real humanoid robots under balance and collision avoidance constraints. Experimental results demonstrate that λ-Reachability significantly improves both safe-set boundary classification and safety margin estimation compared to single-step temporal-difference baselines.

MODEL-BASED WHOLE-BODY CONTROL

Make nonlinear dynamics easier to control

Koopman operator methods lift nonlinear robot dynamics into representations where efficient linear control tools can be applied.

Continual Koopman learning pipeline for high-dimensional legged robots

LEARNED LINEAR DYNAMICS

Koopman learning and control

The continual-learning method progressively expands its dataset and latent space, with theoretical analysis showing monotonically converging linear approximation error. Experiments use linear MPC on Unitree G1, H1, A1, and Go2 robots and ANYmal D across multiple terrains.

Continual learningRefine the model as data grows.

Whole-body safetyPair learned linear dynamics with efficient constrained control.

Scaling analysisStudy how data and representation size affect approximation.

Publications

[C95] 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
Learning for Dynamics and Control Conference, 2025
Citation Formats:
```
    
```
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:

[C109] Whole-Body Safe Control of Robotic Systems with Koopman Neural Dynamics
Sebin Jung, Abulikemu Abuduweili, Jiaxing Li and Changliu Liu
IEEE International Conference on Robotics and Automation, 2026
Citation Formats:
```
    
```
Abstract:

Controlling robots with strongly nonlinear, high-dimensional dynamics remains challenging, as direct nonlinear optimization with safety constraints is often intractable in real time. The Koopman operator offers a way to represent nonlinear systems linearly in a lifted space, enabling the use of efficient linear control. We propose a data-driven framework that learns a Koopman embedding and operator from data, and integrates the resulting linear model with the Safe Set Algorithm (SSA). This allows the tracking and safety constraints to be solved in a single quadratic program (QP), ensuring feasibility and optimality without a separate safety filter. We validate the method on a Kinova Gen3 manipulator and a Go2 quadruped, showing accurate tracking and obstacle avoidance.

[U] Scaling Law of Neural Koopman Operators
Abulikemu Abuduweili, Yuyang Pang, Feihan Li and Changliu Liu
arXiv:2602.19943, 2026
Citation Formats:
```
    
```

HUMAN-TO-HUMANOID

Transfer human motion into whole-body robot skills

Human motion supplies intent, coordination, and demonstrations for teleoperation and downstream learning.

01

H2O

A reinforcement-learning motion imitator maps monocular human motion estimates to real-time whole-body humanoid teleoperation.

02

OmniH2O

A universal whole-body controller supports multiple teleoperation interfaces and provides a route from teleoperated demonstrations to learned whole-body skills.

Publications

[C79] Learning Human-to-Humanoid Real-Time Whole-Body Teleoperation
Tairan He, Zhengyi Luo, Wenli Xiao, Chong Zhang, Kris Kitani, Changliu Liu and Guanya Shi
IEEE/RSJ International Conference on Intelligent Robots and Systems, 2024
Citation Formats:
```
    
```

[C82] OmniH2O: Universal and Dexterous Human-to-Humanoid Whole-Body Teleoperation and Learning
Tairan He, Zhengyi Luo, Xialin He, Wenli Xiao, Chong Zhang, Weinan Zhang, Kris Kitani, Changliu Liu and Guanya Shi
Conference on Robot Learning, 2024
Citation Formats:
```
    
```

VERSATILE WHOLE-BODY CONTROL

One body, many modes, real hardware

HOVER unifies multiple command modes in one policy; ASAP explicitly learns from real rollouts to reduce the mismatch between simulated and physical dynamics.

ONE POLICY, MANY MODES

HOVER

HOVER uses full-body motion imitation as a common abstraction for multiple command modes, enabling seamless transitions among navigation and manipulation behaviors without retraining a separate policy for each mode.

Project page

ALIGN SIMULATION AND REALITY

ASAP

ASAP pretrains a motion policy in simulation, learns a delta action model from real-world rollouts, then integrates that model into simulation for policy fine-tuning.

Project page

Publications

[C88] HOVER: Versatile Neural Whole-Body Controller for Humanoid Robots
Tairan He, Wenli Xiao, Toru Lin, Zhengyi Luo, Zhenjia Xu, Zhenyu Jiang, Changliu Liu, Guanya Shi, Xiaolong Wang, Linxi Fan and Yuke Zhu
IEEE International Conference on Robotics and Automation, 2025
Citation Formats:
```
    
```

[C97] ASAP: Aligning Simulation and Real-World Physics for Learning Agile Humanoid Whole-Body Skills
Tairan He, Jiawei Gao, Wenli Xiao, Yuanhang Zhang, Zi Wang, Jiashun Wang, Zhengyi Luo, Guanqi He, Nikhil Sobanbab, Chaoyi Pan and others others
Robotics: Science and Systems, 2025
Citation Formats:
```
    
```
Abstract:

Humanoid robots hold the potential for unparalleled versatility in performing human-like, whole-body skills. However, achieving agile and coordinated whole-body motions remains a significant challenge due to the dynamics mismatch between simulation and the real world. Existing approaches, such as system identification (SysID) and domain randomization (DR) methods, often rely on labor-intensive parameter tuning or result in overly conservative policies that sacrifice agility. In this paper, we present ASAP (Aligning Simulation and Real-World Physics), a two-stage framework designed to tackle the dynamics mismatch and enable agile humanoid whole-body skills. In the first stage, we pre-train motion tracking policies in simulation using retargeted human motion data. In the second stage, we deploy the policies in the real world and collect real-world data to train a delta (residual) action model that compensates for the dynamics mismatch. Then, ASAP fine-tunes pre-trained policies with the delta action model integrated into the simulator to align effectively with real-world dynamics. We evaluate ASAP across three transfer scenarios: IsaacGym to IsaacSim, IsaacGym to Genesis, and IsaacGym to the real-world Unitree G1 humanoid robot. Our approach significantly improves agility and whole-body coordination across various dynamic motions, reducing tracking error compared to SysID, DR, and delta dynamics learning baselines. ASAP enables highly agile motions that were previously difficult to achieve, demonstrating the potential of delta action learning in bridging simulation and real-world dynamics. These results suggest a promising sim-to-real direction for developing more expressive and agile humanoids.

HUMANOID MANIPULATION

From coordinated motion to useful work

These projects combine perception, locomotion, bimanual manipulation, and task-level structure for object interaction and logistics.

01 · VISUAL SIM-TO-REAL

VIRAL

VIRAL trains visual whole-body loco-manipulation policies entirely in simulation and transfers them zero-shot to a Unitree G1. Its teacher–student pipeline combines privileged-state learning with vision-based policy distillation, large-scale simulation, and domain randomization.

Project page Read paper

02 · NEURO-SYMBOLIC PLANNING

NeSyPack

NeSyPack decomposes a logistics-packing task into subtasks, maps them to atomic skills through a symbolic skill graph, and selects skill parameters, robot configurations, and control strategies. The system won first prize at the 2025 ICRA WBCD competition.

Read paper Competition story

Publications

[C110] VIRAL: Visual Sim-to-Real at Scale for Humanoid Loco-Manipulation
Tairan He, Zi Wang, Haoru Xue, Qingwei Ben, Zhengyi Luo, Wenli Xiao, Ye Yuan, Xingye Da, Fernando Castañeda, Shankar Sastry, Changliu Liu, Guanya Shi, Linxi Fan and Yuke Zhu
IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026
Citation Formats:
```
    
```

[W] NeSyPack: A Neuro-Symbolic Framework for Bimanual Logistics Packing
Bowei Li, Peiqi Yu, Zhenran Tang, Han Zhou, Yifan Sun, Ruixuan Liu and Changliu Liu
RSS 2025 Workshop on Benchmarking Robot Manipulation: Improving Interoperability and Modularity, 2025
Citation Formats:
```
    
```
Abstract:

This paper presents NeSyPack, a neuro-symbolic framework for bimanual logistics packing. Our NeSyPack combines data-driven models and symbolic reasoning to build an explainable hierarchical framework that is generalizable, data-efficient, and reliable. It decomposes a task into subtasks via hierarchical reasoning, and further into atomic skills managed by a symbolic skill graph. The graph selects skill parameters, robot configurations, and task-specific control strategies for execution. This modular design enables robustness, adaptability, and efficient reuse—outperforming end-to-end models that require large-scale retraining. Using NeSyPack, our team won the First Prize in the What Bimanuals Can Do (WBCD) competition at the 2025 IEEE International Conference on Robotics & Automation (ICRA).

Safe motion in cluttered environments

Learn where safe stopping remains possible

Predict future safety violations

Koopman learning and control

H2O

OmniH2O

HOVER

ASAP

VIRAL

NeSyPack

Interested in working on Humanoid Robotics?