Cross-Embodiment Motion Planning by Coupling Body-Agnostic World Model with Differentiable Ego-Dynamics
Cross-Embodiment Motion Planning by Coupling Body-Agnostic World Model with Differentiable Ego-Dynamics
This research presents Hourglass, an hourglass-shaped architecture designed to solve motion planning in cluttered environments across different robot embodiments. By factorizing the problem into a body-agnostic world model and per-robot differentiable ego-dynamics, the method pools data from diverse robots while maintaining physical fidelity. The authors report that their generalist policy outperforms imitation-learning and diffusion-policy baselines by 3.8--10.6 percentage points in success rate on seven simulated robots of five locomotion types. Additionally, after three minutes of teleoperated driving for parameter identification, the model achieves zero-shot transfer to a real skid-steer robot unseen during training, successfully navigating indoor scenes.
The paper introduces Hourglass, an architecture that couples a body-agnostic world model with differentiable ego-dynamics to enable cross-embodiment motion planning. It claims success rate improvements of 3.8--10.6 percentage points over baselines on seven simulated robots and zero-shot transfer to a real skid-steer robot after three minutes of teleoperated driving for parameter identification.
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.
- Environment
- SimulationOpen source S5
- Control
- TeleoperatedOpen source S6
Reported numbers
Reported duration
3 minutes
View original evidence
r within 1.4--2.8 points without an online map. After three minutes of teleoperated driving for identification, the synthetic-data-trained model transfers zero-shot to a real skid-steer robot unseen in training, navigating unseen indoor scenes.
Open source S6locomotion types covered
5
View original evidence
orld model, and imagination training through each robot's template makes the policy body-aware. On seven simulated robots of five locomotion types, the generalist policy outperforms imitation-learning and diffusion-policy baselines on the same world model by 3.8--10.6 percentage points in success rate and matches per-robot specialists without dedicated training, while approaching its parameter-conditioned MPPI teache
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
orld model, and imagination training through each robot's template makes the policy body-aware. On seven simulated robots of five locomotion types, the generalist policy outperforms imitation-learning and diffusion-policy baselines on the same world model by 3.8--10.6 percentage points in success rate and matches per-robot specialists without dedicated training, while approaching its parameter-conditioned MPPI teache
Open source S5
r within 1.4--2.8 points without an online map. After three minutes of teleoperated driving for identification, the synthetic-data-trained model transfers zero-shot to a real skid-steer robot unseen in training, navigating unseen indoor scenes.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org