Research map / Reinforcement learning / Visual and representation RL
Gentle Manipulation Policy Learning via Demonstrations from VLM Planned Atomic Skills
Jiayu Zhou, Qiwei Wu, Jian Li, Zhe Chen, Xiaogang Xiong, Renjing Xu
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 |
|---|---|---|---|
| use atomic-skill VASK demonstrations instead of human teleoperated demonstrations (ablation) | VT-DP · Long-horizon sim tasks, Object Stack | success rate: 0.30 -> 0.90 | Improved |
| use atomic skills instead of direct VLM waypoint planning (ablation) | VASK · Long-horizon sim tasks, Cabinet and Place | success rate: 0.00 -> 0.63 | Improved |
| use atomic skills instead of direct VLM waypoint planning (ablation) | VASK · Long-horizon sim tasks, Open and Place | average contact force (N): 0.482 -> 0.170 | Improved |
| add force-aware penalty to RL reward (ablation) | atomic skills · Contact force over ten trials per task | contact force | Improved |
| add tactile point cloud to visual point cloud (ablation) | VT-DP · Average over 4 tasks | success rate: 0.58 -> 0.79 | Improved |
| add tactile input to RGB policy (ablation) | DP (RGB) · Average over 4 tasks | success rate: 0.26 -> 0.28 | No clear effect |
| use point clouds instead of RGB (ablation) | diffusion policy · Average over 4 tasks | success rate: 0.26 -> 0.58 | Improved |
| use larger/stronger VLM planner (Doubao-pro vs Qwen2.5-7b) (ablation) | VASK planner · Trajectory generation, average | success rate: 0.033 -> 0.713 | Improved |
Closest papers by meaning
- VLBiMan: Vision-Language Anchored One-Shot Demonstration Enables Generalizable Bimanual Robotic Manipulation (ICLR 2026 · similarity 0.90)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- AtomicVLA: Unlocking the Potential of Atomic Skill Learning in Robots (CVPR 2026 · similarity 0.89)
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning (ICML 2025 · similarity 0.89)
- RDD: Retrieval-Based Demonstration Decomposer for Planner Alignment in Long-Horizon Tasks (NeurIPS 2025 · similarity 0.88)
- ForceVLA: Enhancing VLA Models with a Force-aware MoE for Contact-rich Manipulation (NeurIPS 2025 · similarity 0.88)
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.
