Harness Scaling introduces a new scaling dimension for agentic reinforcement learning by training models across diverse harness configurations instead of a single fixed harness. The approach constructs a factorized harness space, selects a compact training subset with balanced factorial coverage, and organizes cross-harness training to combine global coverage with controlled variation for the same task. Experiments across multiple benchmarks, model scales, and RL algorithms show that Harness Scaling consistently improves generalization to unseen white-box configurations and enables effective transfer to unseen external black-box harnesses.
By Fei Bai, Huatong Song, Chuan Hao, Ran Tao, Jian Yang, Bryan Dai, Wayne Xin Zhao, Ji-Rong Wen
arXiv:2608. 19684v1 Announce Type: new Abstract: Recent studies investigate how to leverage pre-collected datasets to improve the policy performance and sample efficiency of RL.
By Tanachai Anakewat, Takayuki Osa, Tatsuya Harada
The paper introduces the Agentic Compositional Generalization hypothesis, suggesting that reinforcement learning (RL) primarily refines high‑level decision‑making behaviors that orchestrate pre‑trained low‑level skills, rather than teaching new domain‑specific skills from scratch. It proposes River, a training recipe that enhances reward quality by filtering low‑quality synthetic environments and adding process‑level behavior regularization. Using River, RL‑trained agents outperform other open‑source 8B models on four terminal‑agent benchmarks, achieving significant gains with fewer than 30% of the training environments.
By Yihang Yao, Bo Pang, Xuan Phi Nguyen, Ding Zhao, Shafiq Joty, Semih Yavuz
arXiv:2512. 09706v2 Announce Type: replace Abstract: The paradigm of agentic AI is shifting from engineered complex workflows to post-training native models.
By Kaichen He, Zihao Wang, Muyao Li, Anji Liu, Yitao Liang
arXiv:2502. 19544v3 Announce Type: replace Abstract: Leveraging offline data is a promising way to improve the sample efficiency of online reinforcement learning (RL).
By Yi Zhao, Aidan Scannell, Wenshuai Zhao, Yuxin Hou, Tianyu Cui, Le Chen, Dieter B\"uchler, Arno Solin, Juho Kannala, Joni Pajarinen
DaCe-DT is an offline multi‑task reinforcement learning framework that tackles data‑level bottlenecks by introducing length‑gated prompt masking, retrieval‑augmented prompt construction, and value‑adaptive return calibration. These mechanisms adapt prompts to diverse task complexities, ensure semantic relevance of prompt segments, and correct misleading supervision from fragmented trajectories. Experiments on Meta‑World show that DaCe‑DT outperforms state‑of‑the‑art methods, improving performance by 11.73% on optimal datasets and 13.34% on suboptimal datasets.
By Xinfei Wang, Shanchen Pang, Chenhao Zhang, Shudong Wang, Wenhao Ji, Haiyuan Gui, Meng Han, Xiaojian Liao
arXiv:2603.02935v2 Announce Type: replace
Abstract: Offline meta-reinforcement learning seeks to learn a policy that generalizes to new related tasks online. Context-based methods infer a task repres...
By Mohammadreza Nakheai, Aidan Scannell, Kevin Luck, Joni Pajarinen
arXiv:2608. 16798v1 Announce Type: cross Abstract: Agent harnesses have substantially improved performance on long-horizon tasks by coordinating agent interactions with the environment.
By Huatong Song, Fei Bai, Ming Yang, Renyuan Li, Jia Deng, Jujie He, Zhange Zhang, Daixuan Cheng, Yan Xing, Qi Yun, Xuxing Chen, Danyang Li, Feng Chang, Chuan Hao, Ran Tao, Jian Yang, Bryan Dai, Wayne Xin Zhao, Mingjie Tang, Ji-Rong Wen
RoboGene is an agentic framework designed to generate diverse and informative manipulation tasks for single-arm, dual-arm, and mobile robots. It employs diversity-driven sampling, agentic self-reflection for task quality assessment, and human-verified feedback memory to validate and refine tasks. The framework produced 18,000 real-world trajectories and introduced new metrics for task feasibility and diversity, outperforming models like GPT‑4o and Gemini 2.5 Pro in training VLA models that achieve higher success rates and stronger generalization.
By Yixue Zhang, Kun Wu, Zhi Gao, Zhen Zhao, Pei Ren, Zhiyuan Xu, Fei Liao, Xinhua Wang, Shichao Fan, Di Wu, Qiuxuan Feng, Meng Li, Zhengping Che, Chang Liu, Jian Tang
The paper argues that zero‑shot generalization in robotics is limited by a lack of diverse training data, unlike digital AI where crowd‑sourced data drives progress. It defines zero‑shot generalization as coverage of the deployment distribution and shows that adding more samples from the same source only marginally reduces the unseen region, whereas new contributors significantly expand it. The authors propose a shared latent space for robot actions, combining multi‑modal alignment, generative augmentation, and noisy‑label learning, and suggest building a community of developers around a single standard robot, Origami, to generate diverse data for testing zero‑shot success.
By Rafael Felix, Flavia Yanase
The paper introduces Trajectory-guided Joint Policy Optimization (TJPO), a reinforcement‑learning framework that explicitly encourages diversity in the trajectories of large‑language‑model agents. By defining task‑specific trajectory descriptors, TJPO measures and optimizes diversity as a set‑level function, avoiding population‑based training. Experiments on Sokoban and ALFWorld demonstrate that TJPO yields diverse, interpretable behaviors while preserving strong task performance.
By Huaiyu Fu, Heng Cao, Hao Wang, Jian Ya, Tao Chen
arXiv:2610.15969v1 Announce Type: new
Abstract: In multi-agent reinforcement learning, exposing a cooperator to partners with diverse strategies during training is key to better collaboration at test...
By Paul Teiletche, Kanghyun Ryu, Kyungjoon Park, Dhruva Tirumala, Markus Wulfmeier, Negar Mehr