Humanoid Dodgeball: Learning Fast Dynamic Obstacle Avoidance Motions for Humanoids

Aoqian Zhang; Chunzheng Wang; Haoran Chen; Annan Tang; Cheng Xiang; Fan Shi; Shuzhi Sam Ge

cs.ROOct 29, 2026

Humanoid Dodgeball: Learning Fast Dynamic Obstacle Avoidance Motions for Humanoids

Authors: Aoqian Zhang, Chunzheng Wang, Haoran Chen, Annan Tang, Cheng Xiang, Fan Shi, Shuzhi Sam Ge

Organizations: Department of Electrical and Computer Engineering, National University of Singapore, Singapore. · Shenzhen EngineAI Robotics Technology Co., Ltd., China. · School of Information and Communication Engineering, Dalian University of Technology, China.

Abstract

Humanoids have become increasingly agile and capable in real-world environments. To operate safely in dynamic scenarios, they must rapidly evade fast-moving obstacles while maintaining stability and physically feasible whole-body motion. Learning such behaviors remains challenging due to the high-dimensional dynamics of humanoid systems and the need to jointly balance reactive obstacle avoidance, motion naturalness, and control robustness. In this work, we present Humanoid Dodgeball, a reinforcement learning (RL) framework for agile dynamic obstacle avoidance in humanoids. Our method enables highly dynamic whole-body evasive maneuvers against incoming projectiles through threat-aware reward shaping, while multiple direction-specific adversarial motion priors regularize the learned behaviors toward stable and diverse human-like motions. Extensive simulation and hardware experiments demonstrate that the proposed framework substantially improves dynamic obstacle avoidance performance across diverse obstacle speeds and launch distances, while preserving robust and physically plausible humanoid motions.

Figures & tables

Fig. 1: Our dynamic obstacle avoidance policy deployed on the EngineAI PM01 humanoid. (Top) The robot dodges by stepping forward. (Bottom) The robot dodges by jumping backward.

Fig. 2: Overview of Humanoid Dodgeball . During simulation training, predictive and near-field threat for key bodies are computed to characterize incoming projectile hazards and guide threat-aware reward shaping for whole-body evasion. Direction-specific adversarial motion priors, selected according to the root escape direction, regularize the learned behaviors toward diverse and natural human-like dodge motions. When the threat subsides, recovery rewards, including orientation and capture-point objectives, encourage rapid stabilization. During hardware deployment, the trained actor is executed using obstacle positions estimated from the motion-capture system.

Dodge rewards ( $T>0$ )
Name	Expression	Weight
Body escape	$r^{\mathrm{escape}}$ (Eq. 5 )	40.0
Body distance	$r^{\mathrm{distance}}$ (Eq. 7 )	50.0
Root-body escape direction	$r^{\mathrm{root-dir}}$ (Eq. 8 )	60.0
Recovery rewards in the absence of threats ( $T=0$ )
Root velocity	$\exp\left(-\left\\|\bm{v}_{xy}^{\mathrm{root}}\right\\|_{2}^{2}/0.5^{2}\right)$	10.0

TABLE I: Reward functions for policy training and their corresponding weights.

Category	Term	Range
Dynamics	Friction coefficient	$[0.1,\,2.0]$
	Joint $K_{p},K_{d}$ scaling	$[0.8,\,1.2]\times$
	Joint position bias	$[-0.05,\,0.05]~\mathrm{rad}$
	Torso COM offset	$[-0.05,\,0.05]~\mathrm{m}$
Sensor noise	Base angular velocity $\bm{\omega}$	$[-0.2,\,0.2]~\mathrm{rad/s}$
	Base roll/pitch $(\phi,\theta)$	$[-0.25,\,0.25]~\mathrm{rad}$

TABLE II: Domain randomization parameters. All quantities are sampled uniformly within the specified ranges.

Fig. 3: Top: Polar heatmaps of obstacle avoidance success rates across different launch distances and approach directions relative to the robot’s heading. Bottom: Success-rate heatmaps across different obstacle launch speeds and distances. The red contour indicates the 60% success-rate boundary.

		Unitree G1			EngineAI PM01
$\mathrm{TTH}$ (s)	Method	$\bm{SR}$ $(\%)$	$\bm{T_{\mathrm{recovery}}}$ $(\mathrm{s})$	$\bm{{P}_{\mathrm{motor}}}$ $(\mathrm{W})$	$\bm{SR}$ $(\%)$	$\bm{T_{\mathrm{recovery}}}$ $(\mathrm{s})$	$\bm{{P}_{\mathrm{motor}}}$ $(\mathrm{W})$
$<0.4$	Ours	11.8	$1.194\pm 0.356$	$569.6\pm 124.5$	14.2	$\bm{1.197\pm 0.439}$	$586.3\pm 145.9$
	REBot [ 14 ]	4.8	$1.772\pm 0.468$	$645.3\pm 93.5$	8.1	$2.293\pm 0.972$	$830.9\pm 266.6$
	PAC-MAN [ 15 ]	5.9	$\bm{0.825\pm 0.467}$	$\bm{316.7\pm 77.2}$	3.9	$1.788\pm 1.382$	$\bm{335.9\pm 139.7}$
$0.4$ – $0.8$	Ours	64.6	$1.438\pm 0.900$	$587.7\pm 105.7$	55.5	$1.369\pm 0.666$	$590.0\pm 129.3$
	REBot [ 14 ]	10.0	$1.735\pm 0.443$	$618.7\pm 86.1$	9.8	$2.162\pm 0.575$	$834.1\pm 228.0$

TABLE III: Dodge performance under different time-to-hit (

\mathrm{TTH}

) ranges in simulation. Recovery time and average motor power are reported as mean

\pm

standard deviation. Bold indicates the best result for each robot and

\mathrm{TTH}

range.

Fig. 4: Visualization of obstacle collision frequency on the G1 and PM01 humanoids during the simulation evaluation.

Fig. 5: t-SNE visualization of avoidance motions from the evaluation rollouts.

Fig. 6: Ablation studies under different time-to-hit ( $\mathrm{TTH}$ ) conditions. (a) Success rate after removing individual reward terms or replacing the direction-specific AMP discriminators with a single discriminator. (b) Fall rate after removing the capture-point recovery reward or using a single AMP discriminator.

Fig. 7: Hardware demonstrations on the EngineAI PM01 humanoid under diverse projectile conditions. (a) A ball aimed toward the robot’s center is avoided by stepping forward. (b) A projectile directed toward the head is avoided through a coordinated forward step and ducking motion. (c) A ball aimed at the right shoulder is avoided by stepping laterally to the left. (d) A projectile targeting the head is avoided through a rightward step combined with a ducking maneuver. These examples demonstrate that the learned policy produces diverse whole-body evasive strategies in response to different obstacle trajectories and incoming directions.

Dynamic Obstacle Avoidance

Dynamic obstacle avoidance studies how robots plan and control safe motion around moving or changing obstacles.

See all 121 papers

Whole-Body Planning for Humanoids Navigating Confined Spaces via Self-Collision Avoidance References
Aug 10, 2026Carlos Gonzalez, Luis SentisConstrained Motion Planning Humanoid Robot Locomotion
Department of Aerospace Engineering and Engineering Mechanics, The University of Texas at Austin, TX 78712, USA.
DODGER: Safety-Guided Reinforcement Learning for Robot Navigation Among Dynamic Obstacles
Sep 30, 2026Sanghyuk Park, Kwanwoo Lee, Taekyung Kim +2RL for Robotics Control Barrier Functions
Department of Intelligence and Information, Seoul National University, Republic of Korea · Department of Robotics, University of Michigan, Ann Arbor, MI, USA · Department of Mechanical Engineering, Yonsei University, Republic of Korea +1
SWAP: Test-Time Steerable Humanoid Whole-Body Collision Avoidance
Oct 19, 2026Zeji Yi, Guanqi He, Xiaofeng Guo +5Robot Collision Avoidance Humanoid Robot Control
Carnegie Mellon University · Amazon FAR

Aug 10, 2026J/K move · Enter open · S save

Whole-Body Planning for Humanoids Navigating Confined Spaces via Self-Collision Avoidance References

Carlos Gonzalez, Luis Sentis

Department of Aerospace Engineering and Engineering Mechanics, The University of Texas at Austin, TX 78712, USA.

Constrained Motion Planning Humanoid Robot Locomotion Kinodynamic Motion Planning Robot Motion Planning Robot Collision Avoidance

Open paper AskPDF arXiv

Humanoid locomotion in highly confined environments requires navigating dense environmental obstacles and complex self-collision bounds while maintaining multi-contact dynamic feasibility. Traditional trajectory optimizers frequently struggle in these restricted spaces, as navigating the large collision space with splines on particle abstractions is insufficient and leads to poor local minima. To address this, we propose a three-stage whole-body planning framework that formulates kinematic path planning directly over kinematically reachable rigid-body volumes. By integrating differentiable collision avoidance into a reachability-constrained formulation, our framework synthesizes volume-informed guides that reliably guide a full-order trajectory optimizer over long horizons. We show that these optimized plans serve as high-quality references to train a residual reinforcement learning policy for robust online execution. We validate our approach on the Unitree G1 humanoid across three benchmark testbeds exceeding NIST emergency response standards, achieving restricted confinement ratios (

C_r < 1.5

). Our framework generates feasible trajectories across 12-to-18-second tasks with complex foot and hand contacts where standard baselines fail, while the learned policy successfully tracks these plans under extensive domain randomization in physics simulation.

Abstract

Dynamic Obstacle Avoidance

Explore similar work

Whole-Body Planning for Humanoids Navigating Confined Spaces via Self-Collision Avoidance References

DODGER: Safety-Guided Reinforcement Learning for Robot Navigation Among Dynamic Obstacles

SWAP: Test-Time Steerable Humanoid Whole-Body Collision Avoidance