Research map / Reinforcement learning / RL from feedback
N2M: Bridging Navigation and Manipulation by Learning Pose Preference from Rollout
Kaixin Chai, Hyunjun Lee, Joseph J Lim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn policy-aware pose preference from rollouts instead of reachability | N2M · RoboCasa CloseDrawer (DP) | success rate: 0.16 -> 0.55 | Improved |
| learn policy-aware pose preference from rollouts instead of reachability | N2M · RoboCasa CloseDrawer (BC Transformer) | success rate: 0.15 -> 0.56 | Improved |
| use N2M instead of Mobi-pi | N2M · RoboCasa CloseDrawer (BC Transformer) | success rate: 0.09 -> 0.56 | Improved |
| use N2M instead of Mobi-pi | N2M · RoboCasa CloseDrawer | inference time: 273.52 s -> 0.07 s | Improved |
| add viewpoint augmentation (ablation) | N2M · PnPCounterToCab (20 rollouts) | success rate | Improved |
| increase number of augmented viewpoints M (ablation) | N2M · PnPCounterToCab | success rate: saturates near M=300 | Improved |
| swap PointBERT for PointNet++ or drop pretrained weights (ablation) | N2M · PnPCounterToCab | success rate | No clear effect |
| use PointTransformerV3 encoder (ablation) | N2M · PnPCounterToCab | success rate | Hurt |
| increase number of training scenes (0 to 5) | N2M · PnPCounterToCab unseen scene | success rate | Improved |
| increase training rollouts to 10-20 | N2M · PnPCounterToCab | success rate: matches oracle at 10, surpasses at 20 | Improved |
| model pose distribution with GMM (K=2) for multimodal tasks | N2M · Close Drawer | pose distribution fit | Improved |
Closest papers by meaning
- Exo2EgoPolicy: Pose-Aligned Cross-View Policy Learning (ECCV 2026 · similarity 0.89)
- Pose6DAug: Physically Plausible Multi-View Object Swapping for Robot Data Augmentation (NeurIPS 2026 · similarity 0.86)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.86)
- MoManipVLA: Transferring Vision-language-action Models for General Mobile Manipulation (CVPR 2025 · similarity 0.86)
- AC-DiT: Adaptive Coordination Diffusion Transformer for Mobile Manipulation (NeurIPS 2025 · similarity 0.86)
- Learning Spatial-Aware Manipulation Ordering (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
