Research map / Vision-language and multimodal / Multimodal LLMs
Towards Spatial Trace with Reasoning in Vision-Language Models for Robotics
Enshen Zhou, Yibo Li, Jingkun An, Jiayuan Zhang, Shanyu Rong, Mengzhen Liu, Yi Han, Yuheng Ji, Huajie Tan, Jiawei He, Pengwei Wang, Zhongyuan Wang and 3 more
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 |
|---|---|---|---|
| add 2D data to SFT data recipe (ablation) | RoboTracer-2B SFT · Q-Spatial | SR: 51.49 -> 69.61 | Improved |
| add 3D data to SFT data recipe (ablation) | RoboTracer-2B SFT · Q-Spatial | SR: 33.52 -> 69.61 | Improved |
| add video data to SFT data recipe (ablation) | RoboTracer-2B SFT · ShareRobotBench | Discrete Frechet Distance: 0.4376 -> 0.1605 | Improved |
| add regression-supervised scale decoder instead of no scale supervision (ablation) | RoboTracer-2B SFT · Q-Spatial | SR: 53.47 -> 69.61 | Improved |
| use regression instead of next-token prediction for scale supervision (ablation) | RoboTracer-2B SFT · Q-Spatial | SR: 57.43 -> 69.61 | Improved |
| model points as (u,v,d) instead of (x,y,z) (ablation) | RoboTracer-2B SFT · ShareRobotBench | Discrete Frechet Distance: 0.2426 -> 0.1605 | Improved |
| model points as (u,v,d) instead of (x,y,z) (ablation) | RoboTracer-2B SFT · TraceSpatial-Bench | SR: 30 -> 31 | Improved |
| add universal spatial encoder before RFT (ablation) | RoboTracer-2B RFT · TraceSpatial-Bench | SR: 36 -> 39 | Improved |
| add RFT after SFT | RoboTracer-2B · TraceSpatial-Bench | SR: 31 -> 39 | Improved |
| add metric-sensitive process reward in RFT (ablation) | RoboTracer-RFT · TraceSpatial-Bench | overall SR: 33 -> 39 | Improved |
| add camera intrinsics and absolute depth input to RFT model | RoboTracer-RFT · TraceSpatial-Bench | overall SR: 39 -> 45 | Improved |
| add noise to depth input | RoboTracer-RFT · TraceSpatial-Bench | overall SR: 45 -> 44 | No clear effect |
Closest papers by meaning
- RoboRefer: Towards Spatial Referring with Reasoning in Vision-Language Models for Robotics (NeurIPS 2025 · similarity 0.91)
- TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policies (ICLR 2025 · similarity 0.90)
- RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics (CVPR 2025 · similarity 0.90)
- MindJourney: Test-Time Scaling with World Models for Spatial Reasoning (NeurIPS 2025 · similarity 0.90)
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 2026 · similarity 0.89)
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · 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.
