AAA games offer a rich testbed for evaluating agent perception, reasoning, and control. However, evaluating whether agents can complete tasks in AAA games is challenging and often human-annotated because task outcomes are often difficult to verify automatically, let alone applying it for Reinforcement Learning (RL). We...
Jia-Jun Chen, Shi-Cheng Fang, Jia-Hao Lu et al.· 0 citations
A recurring failure mode in deployed AI lies at the boundary between the state a task depends on and the model-visible context available when the system acts. In such cases, what looks like weak reasoning can instead reflect failed access: relevant user state, repository state, tool traces, sensor readings, or environm...
Existing benchmarks for GUI agents primarily focus on increasing task difficulty by extending task horizons. Besides, enabling agents to explore unfamiliar environments and successfully complete tasks within them is also a critical yet challenging problem. We introduce LatentGUIWorld, a benchmark of six tasks in three...
Jia-Hao Lu, Yi-Ze Shen, Chuan-Yuan Tan et al.· 0 citations
Real-world embodied agents often pursue independent objectives within a shared physical environment, where their actions can alter the conditions faced by others. Existing benchmarks, however, typically assume shared goals or explicitly prescribed interaction protocols, leaving such emergent physical coupling underexpl...
Jie Yang, Jia-Jun Chen, Jia-Zheng Zhou et al.· arXiv.org· 0 citations
This work introduces MARCO, an evaluator-grounded reinforcement-learning framework that trains molecular editors on bounded proposal--feedback--revision trajectories and aggregates shaped turn rewards into an undiscounted trajectory return for group-relative policy optimization.
Shi-Cheng Fang, Yu-Xin Wang, Zhuo Yang et al.· arXiv.org· 0 citations
Objective-wise Reconciled Policy Gradient (ORPG), which constructs a separate clipped policy objective for each reward and reconciles the resulting gradients into one policy update, achieves the highest average full-budget accuracy and three-budget hypervolume among the compared methods.
Shi-Cheng Fang, Yi-Wen Zhao, Wen-Bo Tian et al.· 0 citations
A paired ablation that removes explicit scientific guidance while preserving the repository and executable engineering context shows that scientific knowledge is not uniformly beneficial: well-grounded information can constrain repair and improve average performance and token efficiency, whereas poorly aligned guidance...
Zhi-Peng Xu, Jia-Hao Lu, Yi-Ning Zheng et al.· 4 citations
The Embodied Task Agent is introduced, a new paradigm for extending digital agents into the physical world, and OpenETA is released as its open-source implementation, which provides replaceable Planners, composable Tools and Skills, auditable memory, replayable trajectories, and common interfaces for simulation and rea...
Yi-Tong Chen, Zezheng Huai, Si-Xian Li et al.· 5 citations· ⚡1
EvoCUA-1.5 extends self-evolving computer-use agents from offline experience learning to online reinforcement learning, where policies interact with executable sandbox environments and improve from verifiable task outcomes and provides a practical framework for scaling online RL in multi-turn computer-use agents.
Mianqiu Huang, Taofeng Xue, Chong Peng et al.· 1 citation
Applications in materials analysis, molecule design, and protein or antibody screening, together with experiments on scientific reading, idea generation, molecule generation, and antibody screening, show that SCION outperforms existing autonomous research-agent baselines, especially in decomposition, verification, refi...
Y. Zheng, Yuxin Wang, Jiahao Lu et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.