Reproduce “DynaMOMA: Instantaneous Prediction of Grasp Poses for Mobile Manipulation of Dynamic Objects” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhinan Yu, Junyan Xu, Jiazhao Zhang, Zheng Qin, Yijie Tang, Yuhang Huang, Yihan Cao, Zhiyuan Yu and 4 more
DynaMOMA predicts short-horizon trajectories of feasible grasp poses to guide a mobile robot’s coordinated base-and-arm movement toward moving objects. Its policy blends predicted poses with the current grasp observation according to distance, supporting feedforward interception rather than reaction to the visible pose alone. In simulated Chasing, it achieves 91.5% GraspSR versus ReachMM’s 6.0%, a pose-match metric, not pickup success.
Key insights
- Instantaneous and temporally consistent grasp trajectory prediction is crucial for intercepting dynamic objects, going beyond simple object motion forecasting.
- Anchor-based diffusion effectively captures diverse and multi-modal grasp trajectory patterns, providing more stable and reliable guidance compared to basic extrapolation methods.
- An adaptive, anticipation-guided reward function allows the whole-body control policy to balance coarse approach with precise, feedforward control, essential for successful dynamic grasping.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.