Research map / Reinforcement learning / RL from feedback
ReGRPO: Reflection-Augmented Policy Optimization for Tool-Using Agents
Binjie Zhang, Mike Zheng Shou
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 structured reflection-aware GRPO RL on top of RoT SFT (ablation) | MAT-Qwen2-VL-7B · GTA | AnsAcc: 58.59 -> 67.66 | Improved |
| add structured reflection-aware GRPO RL on top of RoT SFT (ablation) | MAT-Qwen2-VL-7B · GAIA | AnsAcc: 19.03 -> 23.35 | Improved |
| replace MM-Traj SFT data with RoT reflective data (ablation) | MAT-AGENT · GTA | AnsAcc: 53.85 -> 58.59 | Improved |
| replace MM-Traj SFT data with RoT reflective data (ablation) | MAT-AGENT · GAIA | AnsAcc: 16.97 -> 19.03 | Improved |
| add verifier distillation to RoT SFT (ablation) | RoT SFT · GTA | AnsAcc: 58.59 -> 59.72 | Improved |
| add verifier distillation to RoT SFT (ablation) | RoT SFT · GAIA | AnsAcc: 19.03 -> 19.84 | Improved |
| add reflection tokens (free-form) to GRPO (ablation) | GRPO-only · GTA | AnsAcc: 64.51 -> 65.34 | Improved |
| add reflection tokens (free-form) to GRPO (ablation) | GRPO-only · GAIA | AnsAcc: 18.92 -> 21.38 | Improved |
| use structured schema reflection (ReGRPO) instead of free-form reflection (ablation) | GRPO + free-form reflection · GTA | AnsAcc: 65.34 -> 67.66 | Improved |
| use structured schema reflection (ReGRPO) instead of free-form reflection (ablation) | GRPO + free-form reflection · GAIA | AnsAcc: 21.38 -> 23.35 | Improved |
| add optional verifier reward to ReGRPO (ablation) | ReGRPO core · GTA | AnsAcc: 67.66 -> 68.49 | Improved |
| add optional verifier reward to ReGRPO (ablation) | ReGRPO core · GAIA | AnsAcc: 23.35 -> 24.01 | Improved |
Closest papers by meaning
- RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- Multi-modal Agent Tuning: Building a VLM-Driven Agent for Efficient Tool Usage (ICLR 2025 · similarity 0.88)
- Reflective Prompted Policy Optimization: Trajectory-Grounded Revision and Salience Bias (NeurIPS 2026 · similarity 0.88)
- AgentBrew: Offline Tool-Use Agent Learning from Raw Real-World Trajectories (NeurIPS 2026 · similarity 0.88)
- Joint Optimization of Tool Creation and Use for Large Language Model Agents (NeurIPS 2026 · similarity 0.88)
- Democratizing Tool Learning with Environments Fully Simulated by a Free 8B Language Model (NeurIPS 2026 · 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.
