Research map / Vision-language and multimodal / Multimodal LLMs
From Passive Observer to Active Critic: Reinforcement Learning Elicits Process Reasoning for Robotic Manipulation
Yibin Liu, Yaxing Lyu, Daqi Gao, Zhixuan Liang, Weiliang Tang, Shilong Mu, Xiaokang Yang, Mingyu Ding, Yao Mu
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 GRPO RL stage after SFT (ablation) | SFT-only PRIMO · PRIMO Bench (7-split average) | MRA: 79.35 -> 85.28 | Improved |
| add SFT stage before RL (ablation) | RL-only PRIMO · PRIMO Bench (7-split average) | MRA: 76.72 -> 85.28 | Improved |
| add RL after SFT on Real Humanoid cross-environment (ablation) | SFT-only PRIMO · PRIMO Bench (Real Humanoid) | MRA: 67.30 -> 72.32 | Improved |
| apply SFT to base model (ablation) | Qwen2.5-VL-7B · PRIMO Bench (7-split average) | MRA: 67.46 -> 79.35 | Improved |
| apply SFT only (ablation) | Qwen2.5-VL-7B · RoboFail | accuracy (%): 57.6 -> 51.0 | Hurt |
| apply RL (GRPO) after SFT (ablation) | PRIMO (SFT) · RoboFail | accuracy (%): 51.0 -> 63.0 | Improved |
| add SFT before RL on failure detection (ablation) | PRIMO (RL) · RoboFail | accuracy (%): 63.0 -> 67.0 | Improved |
| add initial state image to current image (ablation) | I_curr only · PRIMO Bench (avg MAE) | MAE: 59.50 -> 46.50 | Improved |
| replace current-only input with video sequence (ablation) | I_curr only · PRIMO Bench (avg MAE) | MAE: 59.50 -> 36.61 | Improved |
| add both boundary images to video input (ablation) | V_seq only · PRIMO Bench (avg MAE) | MAE: 36.61 -> 31.43 | Improved |
| add video sequence to boundary-state images on Behavior (ablation) | I_init + I_curr · Behavior | MAE: 49.59 -> 22.73 | Improved |
| add initial image to V_seq + I_curr on AgiBot (ablation) | V_seq + I_curr · AgiBot | MAE: 25.04 -> 29.39 | Hurt |
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.
- Robobrain 7B — RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete (CVPR 2025) · better on 1 of 1 shared result
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
- SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- RoboProcessBench: Benchmarking Process-Aware Understanding in Visual-Language Robotic Manipulation (NeurIPS 2026 · similarity 0.89)
- RoboAlign-R1: Distilled Multimodal Reward Alignment for Robot Video World Models (NeurIPS 2026 · similarity 0.88)
- AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation (ICLR 2025 · similarity 0.87)
- Chain-of-Action: Trajectory Autoregressive Modeling for Robotic Manipulation (NeurIPS 2025 · similarity 0.87)
- SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation (ICML 2025 · similarity 0.87)
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.
