Reinforcement Learning

写在前面

Sim RL, real world RL, RL with Human-in-the-loop, 我们寻找reward，为科研也为生活。

图源: https://arxiv.org/pdf/2508.08189

RL & VLA: A Review

What Can RL Bring to VLA Generalization? An Empirical Study

• paper link: https://arxiv.org/abs/2505.19789
• time: 25/06
• core idea:
• reading note:
• 推荐指数：

MoRE(No-env offline)

Unlocking Scalability in Reinforcement Learning for Quadruped Vision-Language-Action Models

• paper link: https://arxiv.org/abs/2503.08007
• time: 25/06
• core idea:
• reading note:
• 推荐指数：

ReinboT(No-env offline)

Amplifying Robot Visual-Language Manipulation with Reinforcement Learning

• paper link: https://arxiv.org/abs/2503.08007
• time: 25/03
• core idea:
• reading note:
• 推荐指数：

OctoNav(Env-sim online)

Towards Generalist Embodied Navigation

• paper link: https://arxiv.org/pdf/2506.09839
• time: 25/06
• core idea:
• reading note:
• 推荐指数：

ConRFT(Env-real online, RSS)

A Reinforced Fine-tuning Method for VLA Models via Consistency Policy

• paper link: https://arxiv.org/abs/2502.05450
• time: 25/02
• core idea:
• reading note:
• 推荐指数：

RLGD(Env-real online, RSS)

• paper link: https://arxiv.org/abs/2412.09858
• time: 25/06
• core idea:
• reading note:
• 推荐指数：

Policy Agnostic(Env-real online)

Offline RL and Online RL Fine-Tuning of Any Class and Backbone

• paper link: https://arxiv.org/abs/2412.09858
• time: 25/06
• core idea:
• reading note:
• 推荐指数：

GRAPE (test time)

Generalizing Robot Policy via Preference Alignment

• paper link: https://arxiv.org/pdf/2411.19309
• time: 24/11
• core idea:
• reading note:
• 推荐指数：

BAIR (science, real) Precise and Dexterous Robotic Manipulation via Human-in-the-Loop Reinforcement Learning

• paper link: https://arxiv.org/abs/2410.21845

• time: 2024/10

• core idea:

• reading note: omg读会读多了，期刊真的不太好读啊…正文22页我哭死

• 推荐指数：