Research map / Reinforcement learning / Visual and representation RL
EVOLVING ROLLOUTS: Harnessing Historical Experience for Web Agent Evolution in Reinforcement Learning
Sinuo Wang, WANG PIAOHONG, Tianrui Qin, Maojia Song, Qianben Chen, Qiexiang Wang, Gengze Zhou, Zeyu Zhang, He Zhu, Dingfeng Shi, Yutong Xie, Minghao Liu and 6 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 |
|---|---|---|---|
| co-evolve experience repository with the policy during RL instead of keeping it static (ablation) | EVOLVING ROLLOUTS · GAIA, xBench, HLE average | average Pass@1: 31.3 -> 33.9 (Qwen3-8B); 30.6 -> 31.2 (Qwen3-4B) | Improved |
| add experience context with static repository on top of SFT+RL (ablation) | Qwen3-4B · GAIA, xBench, HLE average | average Pass@1: 28.9 -> 30.6 | Improved |
| evolve experience while freezing the policy (ablation) | Qwen3-4B · GAIA, xBench, HLE average | average Pass@1: 28.9 -> 23.0 | Hurt |
| pair fixed step-5 policy with later experience snapshots (instead of synced policy) (ablation) | EVOLVING ROLLOUTS · GAIA at step 31 | GAIA score: 44.7 -> 36.9 | Hurt |
| add experience context at ReAct stage (no training) (ablation) | Qwen3-4B ReAct · GAIA, xBench, HLE average | average Pass@1: 22.8 -> 23.0 | No clear effect |
| add experience context after SFT (ablation) | Qwen3-4B SFT · GAIA, xBench, HLE average | average Pass@1: 25.3 -> 26.3 | Improved |
| add experience evolution during RL (ablation) | Qwen3-4B SFT+RL · GAIA | Pass@1: 41.7 -> 44.7 | Improved |
| add experience evolution during RL (ablation) | Qwen3-4B · xBench-DeepSearch | Pass@1: static 38.0 -> evolving 36.0 | Mixed |
| add experience evolution during RL (ablation) | Qwen3-8B · xBench-DeepSearch | Pass@1: 44.0 -> 44.0 | No clear effect |
Closest papers by meaning
- WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning (ICLR 2025 · similarity 0.88)
- EvoTest: Evolutionary Test-Time Learning for Self-Improving Agentic Systems (ICLR 2026 · similarity 0.87)
- StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction (NeurIPS 2026 · similarity 0.87)
- WebPilot: A Versatile and Autonomous Multi-Agent System for Web Task Execution with Strategic Exploration (AAAI 2025 · similarity 0.87)
- Stratified GRPO: Handling Structural Heterogeneity in Reinforcement Learning of LLM Search Agents (ICML 2026 · similarity 0.86)
- WebGym: Scaling Training Environments for Long-Horizon Visual Web Agents with Realistic Tasks (CVPR 2026 · similarity 0.86)
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.
