网站内容建设的建议黑马程序员视频库

网站内容建设的建议,黑马程序员视频库,成都旅游团,网站开发硬件工程师待遇主页#xff1a;http://qingkeai.online/ 原文#xff1a;https://mp.weixin.qq.com/s/lfkwxQ-7N2jdVaOFAN5GmQ 随着基于大规模模仿学习的视觉-语言-动作 (VLA) 模型取得显著进展#xff0c;将 VLA与强化学习 (RL)相结合已成为一种极具前景的新范式。该范式利用与环境的试错…主页http://qingkeai.online/原文https://mp.weixin.qq.com/s/lfkwxQ-7N2jdVaOFAN5GmQ随着基于大规模模仿学习的视觉-语言-动作 (VLA) 模型取得显著进展将VLA与强化学习 (RL)相结合已成为一种极具前景的新范式。该范式利用与环境的试错交互或预先采集的次优数据进一步提升机器人的决策与执行能力。本文对该领域的关键论文进行了分类整理涵盖离线RL、在线RL、世界模型、推理时RL及对齐技术。一、 离线强化学习 (Offline RL)离线 RL 预训练的 VLA 模型利用人类演示和自主收集的数据进行学习无需实时环境交互。Q-Transformer: Scalable Offline Reinforcement Learning via Autoregressive Q-Functions链接https://arxiv.org/abs/2309.10150 代码https://github.com/google-deepmind/q_transformerOffline Actor-Critic Reinforcement Learning Scales to Large Models (Perceiver-Actor-Critic)链接https://arxiv.org/abs/2402.05546 代码https://offline-actor-critic.github.io/GeRM: A Generalist Robotic Model with Mixture-of-experts for Quadruped Robot链接https://arxiv.org/abs/2403.13358 代码https://github.com/Improbable-AI/germReinboT: Amplifying Robot Visual-Language Manipulation with Reinforcement Learning链接https://arxiv.org/abs/2505.07395MoRE: Unlocking Scalability in Reinforcement Learning for Quadruped Vision-Language-Action Models链接https://arxiv.org/abs/2503.08007CO-RFT: Efficient Fine-Tuning of Vision-Language-Action Models through Chunked Offline Reinforcement Learning链接https://arxiv.org/pdf/2508.02219Balancing Signal and Variance: Adaptive Offline RL Post-Training for VLA Flow Models (ARFM)链接https://arxiv.org/pdf/2509.04063二、 在线强化学习 (Online RL)通过在环境中的试错交互进一步优化 VLA 模型的性能。1. 仿真环境内 (In Simulator)FLaRe: Achieving Masterful and Adaptive Robot Policies with Large-Scale Reinforcement Learning Fine-Tuning链接https://arxiv.org/abs/2409.16578 代码https://github.com/flare-vla/flarePolicy Agnostic RL: Offline RL and Online RL Fine-Tuning of Any Class and Backbone (PA-RL)链接https://arxiv.org/abs/2412.06685 代码https://pa-rl.github.io/Improving Vision-Language-Action Model with Online Reinforcement Learning (iRe-VLA)链接https://arxiv.org/abs/2501.16664Interactive Post-Training for Vision-Language-Action Models (RIPT-VLA)链接https://arxiv.org/abs/2505.17016 代码https://github.com/OpenHelix-Team/RIPTVLA-RL: Towards Masterful and General Robotic Manipulation with Scalable Reinforcement Learning链接https://arxiv.org/abs/2505.18719 代码https://github.com/vla-rl/vla-rlWhat Can RL Bring to VLA Generalization? An Empirical Study (RLVLA)链接https://arxiv.org/abs/2505.19789 代码https://github.com/S-S-X/RLVLARFTF: Reinforcement Fine-tuning for Embodied Agents with Temporal Feedback链接https://arxiv.org/abs/2505.19767SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning周六上午10点一起聊聊VLA强化学习训练框架SimpleVLA-RL链接https://arxiv.org/pdf/2509.09674 代码https://github.com/SimpleVLA/SimpleVLATGRPO: Fine-tuning Vision-Language-Action Model via Trajectory-wise Group Relative Policy Optimization链接https://arxiv.org/abs/2506.08440 代码https://github.com/TGRPO/TGRPOOctoNav: Towards Generalist Embodied Navigation链接https://arxiv.org/abs/2506.09839 代码https://octonav.github.io/RLRC: Reinforcement Learning-based Recovery for Compressed Vision-Language-Action Models链接https://arxiv.org/pdf/2506.17639 代码https://rlrc-vla.github.io/RLinf: Reinforcement Learning Infrastructure for Agentic AI下周二晚8点和无问芯穹首席研究员林灏一起聊聊具身智能 RL 训练框架 RLinf 的系统设计链接https://arxiv.org/pdf/2509.15965 代码https://rlinf.github.io/RLinf-VLA: A Unified and Efficient Framework for VLARL TrainingVLARL 算法如何设计从零上手 OpenVLA 的强化学习微调实践链接https://arxiv.org/pdf/2510.06710v12. 真实世界 (In Real-World)RLDG: Robotic Generalist Policy Distillation via Reinforcement Learning链接https://arxiv.org/abs/2412.09858 代码https://rldg.github.io/Policy Agnostic RL: Offline RL and Online RL Fine-Tuning of Any Class and Backbone链接https://arxiv.org/abs/2412.06685 代码https://github.com/MaxSobolMark/PolicyAgnosticRLImproving Vision-Language-Action Model with Online Reinforcement Learning链接https://arxiv.org/abs/2501.16664ConRFT: A Reinforced Fine-tuning Method for VLA Models via Consistency Policy链接https://arxiv.org/abs/2502.05450 代码https://github.com/ConRFT/ConRFTVLAC: A Vision-Language-Action-Critic Model for Robotic Real-World Reinforcement Learning链接https://arxiv.org/abs/2509.15937 代码https://github.com/VLAC-VLA/VLACSelf-Improving Embodied Foundation Models (Generalist)链接https://arxiv.org/pdf/2509.15155三、 世界模型 (World Model / Model-Based RL)利用世界模型作为虚拟环境实现低成本、安全的 VLA 策略后训练。World-Env: Leveraging World Model as a Virtual Environment for VLA Post-Training链接https://arxiv.org/abs/2509.24948 代码https://github.com/amap-cvlab/world-envVLA-RFT: Vision-Language-Action Reinforcement Fine-tuning with Verified Rewards in World Simulators链接https://arxiv.org/pdf/2510.00406 代码https://github.com/VLA-RFT/VLA-RFT四、 推理时强化学习 (Test-Time RL)在部署阶段利用预训练的价值函数进行实时优化或纠错。To Err is Robotic: Rapid Value-Based Trial-and-Error during Deployment (Bellman-Guided Retrials)链接https://arxiv.org/abs/2406.15917 代码https://github.com/notmahi/bellman-guided-retrialsSteering Your Generalists: Improving Robotic Foundation Models via Value Guidance (V-GPS)链接https://arxiv.org/abs/2410.13816 代码https://v-gps.github.io/Hume: Introducing System-2 Thinking in Visual-Language-Action Model链接https://arxiv.org/abs/2505.21432 代码https://github.com/Hume-VLA/HumeVLA-Reasoner: Empowering Vision-Language-Action Models with Reasoning via Online Monte Carlo Tree Search链接https://arxiv.org/abs/2509.22643五、 强化学习对齐 (RL Alignment)旨在使 VLA 策略符合人类偏好或安全约束。GRAPE: Generalizing Robot Policy via Preference Alignment链接https://arxiv.org/abs/2411.19309 代码https://github.com/GRAPE-VLA/GRAPESafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning链接https://arxiv.org/abs/2503.03480 代码https://safevla.github.io/六、 其他分类 (Unclassified)RPD: Refined Policy Distillation: From VLA Generalists to RL Experts链接https://arxiv.org/abs/2503.05833总结VLA 与 RL 的结合正处于快速爆发期。将模仿学习的大规模先验与强化学习的自进化能力相结合是通向具身通用人工智能的关键路径。都看到这了点个关注再走吧