Research map / Vision-language and multimodal / Robot action and navigation
Rethinking Intermediate Representation for VLM-based Robot Manipulation
Weiliang Tang, Jialin Gao, Jia-Hui Pan, Gang Wang, Li Erran Li, Yun-Hui Liu, Mingyu Ding, Pheng-Ann Heng, Chi-Wing Fu
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 SEAM semantic-assembly representation instead of OmniManip's keypoint/axis representation | VLM-based manipulation with Qwen3-VL · Real-world UR5 8 tasks (closed-loop) | total success rate: 68.8% -> 83.8% | Improved |
| use SEAM instead of OmniManip in open-loop execution | VLM-based manipulation with Qwen3-VL · Real-world UR5 8 tasks (open-loop) | total success rate: 52.5% -> 63.8% | Improved |
| use SEAM instead of OmniManip on pick up cup/bowl onto dish (open-loop) | VLM-based manipulation with Qwen3-VL · Pick up cup/bowl onto the dish (open-loop) | success rate: 7/10 -> 5/10 | Hurt |
| add RAG-based few-shot part segmentation to localize the teapot opening | SEAM · Fit the lid onto the teapot | qualitative execution | Improved |
| use RAG few-shot segmentation instead of LISA, OV-seg or Grounded SAM | SEAM · Part segmentation inference time (A6000) | time (sec.): 0.88 -> 0.6 vs Grounded SAM | Improved |
| use high-level skill vocabulary (Instruct2Act) instead of low-level primitives | intermediate representation · 33 generated tasks, Qwen3-VL | action-generalizability vs VLM-comprehensibility | Mixed |
| use low-level keypoint representation (ReKep, OmniManip) instead of SEAM | intermediate representation · 33 generated tasks, Qwen3-VL | action-generalizability vs VLM-comprehensibility | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OmniManip — OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints (CVPR 2025) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints (CVPR 2025 · similarity 0.91)
- Vision-Language Foundation Models as Effective Robot Imitators (ICLR 2024 · similarity 0.90)
- PAC Bench: Do Foundation Models Understand Prerequisites for Executing Manipulation Policies? (NeurIPS 2025 · similarity 0.89)
- ManipLLM: Embodied Multimodal Large Language Model for Object-Centric Robotic Manipulation (CVPR 2024 · similarity 0.89)
- UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent (ICML 2025 · similarity 0.89)
- FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies (NeurIPS 2026 · similarity 0.89)
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.
