Learning Agile Humanoid Dodgeball with Onboard Monocular Vision
Learning Agile Humanoid Dodgeball with Onboard Monocular Vision
This research presents a system for humanoid robots to perform agile dodgeball maneuvers using onboard monocular vision. The approach integrates visual threat assessment with whole-body evasion and recovery policies. A three-stage curriculum transitions the robot from privileged inputs to estimator outputs, utilizing analytic projection and perception randomization for training. Results show high success rates in simulation (97.8-100.0% for 5.0 m launches at 6 m/s) and over 85% success on a physical Unitree G1 across more than 15 trials.
The study demonstrates an integrated onboard monocular vision system enabling agile humanoid dodgeball maneuvers in simulation and on a physical Unitree G1, with success rates varying by launch speed and distance.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
throws per condition
1,000
View original evidence
rom stable policy rollouts, and finally replaces the actor's privileged inputs with estimator outputs. Analytic projection and perception randomization provide detector-like training observations without image rendering. Across approach-angle bins in controlled simulation, dodge success ranges from 97.8--100.0% for 5.0 m launches at 6 m/s to 33.2--40.8% for 2.5 m launches at 10 m/s, with 1000 throws per speed--distan
Open source S4trials for single-throw success
>15 trials
Reported trials
View original evidence
ce condition. On a physical Unitree G1, controlled single-throw success exceeds 85% over more than 15 trials. A hardware sequence demonstrates full-azimuth target reacquisition followed by another dodge. These results demonstrate an integrated onboard-monocular-vision system under highly dynamic motions.
Open source S5training throws per condition
1,000
View original evidence
rom stable policy rollouts, and finally replaces the actor's privileged inputs with estimator outputs. Analytic projection and perception randomization provide detector-like training observations without image rendering. Across approach-angle bins in controlled simulation, dodge success ranges from 97.8--100.0% for 5.0 m launches at 6 m/s to 33.2--40.8% for 2.5 m launches at 10 m/s, with 1000 throws per speed--distan
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ce condition. On a physical Unitree G1, controlled single-throw success exceeds 85% over more than 15 trials. A hardware sequence demonstrates full-azimuth target reacquisition followed by another dodge. These results demonstrate an integrated onboard-monocular-vision system under highly dynamic motions.
Open source S5
arXiv:2611.01197v1 Announce Type: new Abstract: Humanoid robots operating in shared, dynamic environments must respond to rapidly approaching objects while maintaining balance. We study dodgeball as a controlled test of this capability, coupling visual threat assessment with whole-body evasion, recovery, and target reacquisition. We develop a system that integrates onboard monocular perception, temporal ball-state an
Open source S2
rom stable policy rollouts, and finally replaces the actor's privileged inputs with estimator outputs. Analytic projection and perception randomization provide detector-like training observations without image rendering. Across approach-angle bins in controlled simulation, dodge success ranges from 97.8--100.0% for 5.0 m launches at 6 m/s to 33.2--40.8% for 2.5 m launches at 10 m/s, with 1000 throws per speed--distan
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org