Tag
StructRL introduces an online reinforcement learning framework that uses structured intermediate rewards from verifiable subtasks to enhance long-horizon vision-language-action tasks, demonstrating superior performance on benchmarks like RoboCasa365 and LIBERO-Long.
VLA-Precision introduces an efficient real-world online reinforcement learning framework using Asymmetric Co-Bootstrapping to improve Vision-Language-Action models, achieving high success rates in precision tasks.
vLLM introduces a native sharded weight transfer engine using Ray Direct Transport (RDT) for efficient weight syncing in large-scale online RL setups, enhancing performance for models like Kimi K2.
This paper proposes Threat-guided Policy-aware Scene Perturbation (TPSP), a method that augments online reinforcement learning for safe autonomous driving by perturbing scenes in a policy-aware, targeted manner to generate high-value safety-critical experiences. Experiments on NAVSIM v2 show improved safety learning efficiency with about 4 million kilometers of simulated driving data.
This paper proposes a novel online reinforcement learning method to improve factuality in reasoning LLMs by designing a reward function that balances factual precision, detail, and relevance, achieving a 23.1 percentage point reduction in hallucination rate on six benchmarks.
EvoCUA-1.5 introduces an online reinforcement learning framework for multi-turn computer-use agents, achieving a 63.2% success rate on OSWorld-Verified and outperforming comparable open-weight models up to 35B parameters through step-level policy optimization and dynamic curriculum learning.
A paper proposing a mechanism for enterprise agents to improve by safely converting messy daily work into learning data, using a data proxy and control layer, with AREAL2.0 demonstrating online RL from real interaction traces.
This paper proposes Hierarchical Advantage-Weighted Behavior Cloning (HABC) for fine-tuning Vision-Language-Action (VLA) policies using online reinforcement learning with sparse binary episode outcomes. HABC separates viability and efficiency objectives via adaptive critic heads and intervention-aware credit assignment, significantly improving success rates on contact-rich bimanual manipulation tasks.
OpenWebRL presents an open framework for training visual web agents using online multi-turn reinforcement learning on real websites, achieving state-of-the-art performance with minimal initial supervision. Their 4B-parameter model outperforms prior open agents and competes with proprietary systems like OpenAI CUA and Gemini CUA.