FlashNeRD: Performance-First Contact-Rich Neural Robot Dynamics
FlashNeRD: Performance-First Contact-Rich Neural Robot Dynamics
This paper presents FlashNeRD, a neural robot dynamics model that addresses key limitations of existing approaches by introducing a parallel streaming architecture, flexible contact encoding, and two-way coupling with simulated objects. Experiments show significant speedups and accuracy improvements in dynamics prediction, policy learning, and planning across multiple robots.
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
speedup
55
View original evidence
FlashNeRD's dynamics model is up to $55\times$ faster than an optimized NeRD
Open source S4speedup
3.8
View original evidence
PPO trains an ANYmal locomotion policy in 34 seconds, $3.8\times$ faster than the analytical simulator
Open source S4speedup
2.7
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
w faster and more accurate dynamics, faster policy learning, and faster inference-time planning. Across three robots, FlashNeRD's dynamics model is up to $55\times$ faster than an optimized NeRD and more accurate over long rollouts. This speedup extends to policy learning, where PPO trains an ANYmal locomotion policy in 34 seconds, $3.8\times$ faster than the analytical simulator and $2.7\times$ faster than an optimi
Open source S4
zed NeRD. With DIAL-MPC, a sampling-based MPC method, the robot climbs all six test platforms where fixed-contact NeRD manages one, at approximately half the analytical simulator's planning time. Cube-reorientation policies trained with FlashNeRD complete within 2% of the simulator-trained policy's target count.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org