Recently, improving the robustness of policies across different environments attracts increasing attention in the reinforcement learning (RL) community. Existing robust RL methods mostly aim to achieve the max-min robustness by optimizing the policy's performance in the worst-case environment. However, in practice, a user that uses an RL policy may have different preferences over its performance across environments. Clearly, the aforementioned max-min robustness is oftentimes too conservative to satisfy user preference. Therefore, in this paper, we integrate user preference into policy learning in robust RL, and propose a novel User-Oriented Robust RL (UOR-RL) framework. Specifically, we define a new User-Oriented Robustness (UOR) metric for RL, which allocates different weights to the environments according to user preference and generalizes the max-min robustness metric. To optimize the UOR metric, we develop two different UOR-RL training algorithms for the scenarios with or without a priori known environment distribution, respectively. Theoretically, we prove that our UOR-RL training algorithms converge to near-optimal policies even with inaccurate or completely no knowledge about the environment distribution. Furthermore, we carry out extensive experimental evaluations in 4 MuJoCo tasks. The experimental results demonstrate that UOR-RL is comparable to the state-of-the-art baselines under the average and worst-case performance metrics, and more importantly establishes new state-of-the-art performance under the UOR metric.
Reinforcement Learning based Condition-oriented Maintenance Scheduling for Flow Line Systems
Center for Cyber Cognitive Intelligence (CCI), Fraunhofer IPA（弗劳恩霍夫制造工程与自动化研究所网络认知智能中心）；Institute of Industrial Manufacturing and Management IFF, University of Stuttgart（斯图加特大学工业制造与管理研究所）
ROAD: Responsibility-Oriented Reward Design for Reinforcement Learning in Autonomous Driving
Global Institute of Future Technology（未来技术全球研究院）；Shanghai Jiao Tong University（上海交通大学）；College of Humanities（人文学院）；Department of Automation（自动化系）；School of Aeronautics and Astronautics（航空宇航学院）
Revealing the Learning Process in Reinforcement Learning Agents Through Attention-Oriented Metrics
Max Planck Institute for Human Cognitive and Brain Sciences（马克斯·普朗克人类认知与脑科学研究所）；Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI)（可扩展数据分析与人工智能中心（ScaDS.AI））
Mila - Québec AI Institute（米拉-魁北克人工智能研究所）；New York University（纽约大学）；Farama Foundation（法拉马基金会）；University of Southampton（南安普顿大学）；Miple；Telekinesis；Google DeepMind（谷歌DeepMind）；Swarm Labs（Swarm实验室）
Modular Object-Oriented Games: A Task Framework for Reinforcement Learning, Psychology, and Neuroscience
Massachusetts Institute of Technology（麻省理工学院）；Center for Brains, Minds and Machines, MIT（麻省理工学院脑、心智与机器中心）；McGovern Institute of Brain Research, MIT（麻省理工学院麦戈文脑科学研究所）
Goal-Oriented Skill Abstraction for Offline Multi-Task Reinforcement Learning
C2DL, Institute of Automation, Chinese Academy of Sciences（C2DL，自动化研究所，中国科学院）；School of Artificial Intelligence, University of Chinese Academy of Sciences（人工智能学院，中国科学院大学）；Tencent AI Lab（腾讯AI实验室）；Tsinghua University（清华大学）
Deep Reinforcement Learning for Delay-Oriented IoT Task Scheduling in Space-Air-Ground Integrated Network
University of Waterloo（滑铁卢大学）；Zhejiang University（浙江大学）；Huazhong University of Science and Technology（华中科技大学）；Central South University（中南大学）；Xidian University（西安电子科技大学）
RTLSeek: Boosting the LLM-Based RTL Generation with Multi-Stage Diversity-Oriented Reinforcement Learning
State Key Lab of Processors, Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所处理器国家重点实验室）；University of Chinese Academy of Sciences（中国科学院大学）；Peking University（北京大学）；CASTEST Co., Ltd.（中科芯测科技有限公司）
CNA: An AI-Oriented Comprehensive Normalized Assessment for Healthy Status and Application to Optimize RRT Strategies by Reinforcement Learning
Chongqing Institute of Green and Intelligent Technology, CAS（中国科学院重庆绿色智能技术研究院）；Chongqing School, University of Chinese Academy of Sciences（中国科学院大学重庆学院）；Department of Anesthesiology, Southwest Hospital, Third Military Medical University (Army Medical University)（陆军军医大学（第三军医大学）西南医院麻醉科）；College of Computer and Information Science, Southwest University（西南大学计算机与信息科学学院）；Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所）
No-Regret Exploration in Goal-Oriented Reinforcement Learning
Facebook AI Research（Facebook 人工智能研究院）；Inria Lille - Nord Europe（法国国家信息与自动化研究所里尔-北欧中心）
许多流行的强化学习问题（例如，迷宫中的导航，一些Atari游戏，山地车）是其随机最短路径（SSP）公式下的情节设置的实例，其中代理必须在最小化累积成本的同时实现目标状态。尽管这种设置很受欢迎，但在一般SSP问题中，探索-开发困境的研究很少，大多数理论文献都集中在不同的问题上（即，固定范围和无限范围）或进行限制性无环SSP假设（即，在一次事件中，任何州都不能访问两次）。在本文中，我们研究了一般的SSP问题，没有假设其动态（一些政策实际上可能永远不会达到目标）。我们介绍了UC-SSP，第一个无遗憾算法在这种情况下，并证明了一个遗憾的边界缩放为$\displaystyle \widetilde{\mathcal{O}}( D S \sqrt{ A D K})$后，$K$事件的任何未知SSP与$S$状态，$A$行动，积极的成本和SSP直径$D$，定义为从任何起始状态到目标的最小预期命中时间。我们通过设计一个新的停止规则来实现这一结果，这样UC-SSP可以在实现目标花费太长时间的情况下中断当前策略，并切换到旨在快速终止事件的替代策略。
QECO: A QoE-Oriented Computation Offloading Algorithm based on Deep Reinforcement Learning for Mobile Edge Computing
Department of Computer Engineering, Sharif University of Technology（计算机工程系，谢赫拉特夫大学）；Department of Electrical Engineering, Sharif University of Technology（电气工程系，谢赫拉特夫大学）
Efficient Exploration in Constrained Environments with Goal-Oriented Reference Path
Information Technology R&D Center, Mitsubishi Electric Corporation（三菱电机株式会社信息技术研发中心）；National Institute of Advanced Industrial Science and Technology (AIST)（日本先进产业科学技术研究所（AIST））；Mitsubishi Electric Research Labs（三菱电机研究实验室）
Reinforcement Learning for Personalized Drug Discovery and Design for Complex Diseases: A Systems Pharmacology Perspective
Hunter College, The City University of New York（纽约城市大学亨特学院）；The Graduate Center, The City University of New York（纽约城市大学研究生中心）；Weill Cornell Medicine, Cornell University（康奈尔大学威尔康奈尔医学院）
