PPO‑HRAP introduces a hybrid regime‑aware policy that blends Proximal Policy Optimization with a volatility‑conditioned regime prior to balance upside participation and drawdown control in trading. The agent uses market and portfolio features, rewards that combine log return, VIX‑conditioned drawdown penalty, exposure deviation, and turnover cost, and outputs a blended action between the PPO actor and the regime‑derived target exposure. In backtests on SPY (2020‑2022) it achieved a 27.62% total return, 8.48% annualized return, and reduced maximum drawdown from 34.10% to 18.47%, while maintaining stable performance across multiple seeds and ranking first on total return and Sharpe ratio in single‑run cross‑asset tests on QQQ and DIA.
By Duong Hien Chi Kien, Thanh Trung Huynh
Beyond Memory: Harnessing Long-Horizon Agents with Explicit Belief States introduces PoS, an inference-time framework that builds and maintains explicit belief states to guide large language model agents. Each belief state combines an estimate of the current world with unresolved task requirements, making clear what the agent still needs to learn and accomplish. PoS validates consistency, monitors task progress to detect Belief Trapping, and tailors recovery to the trapping pattern and unresolved requirements, achieving top performance across four benchmarks with all three LLM backbones.
By Yu Luo, Jiamin Jiang, Yimin Zuo, Xidao Wen, Rongchen Gao, Yongqian Sun, Shenglin Zhang, Guiyang Liu, Cheng Zhang, Fang Situ, Qi Zhou, Dan Pei
The paper introduces MCRI, a four‑dimensional framework for analyzing and evaluating agent skills, and implements it as MCRI‑Eval, a large‑language‑model‑based evaluation method. Using 63,812 public skills from OpenClaw and 58,275 skill‑conditioned executions across BigCodeBench, BFCL‑Fundamental, and Mind2Web, MCRI‑Eval’s scores correlate with community popularity and outperform other methods in downstream ranking agreement. The evaluation also shows that MCRI‑Eval improves top‑1 skill selection by 17.7, 22.8, and 19.6 percentile points on the three benchmarks, offering a valuable pre‑execution signal for prioritizing promising skills before costly evaluation.
By Zongrui Yang, Li Xintong, Runchen Xu, Zhongsheng Wang, Zhedong Lin, Haoyuan Li, Jiamou Liu
The paper argues that evaluating agents as fixed models is insufficient, proposing instead to treat them as configurable systems. Using a new benchmark of four scientific tasks, the authors analyze how five configuration aspects—task information, reasoning, self‑verification, time budget, and backbone model—affect performance, noting that about 54% of outcome variance arises from run‑to‑run differences even with the same settings. The study finds that providing more task information has the strongest impact, while interactions among settings (e.g., extra time only helps with adequate information or model capability) and the choice of verification tools significantly shape agent behavior.
By Luis Wiedmann, Leander Girrbach, Cordelia Schmid, Zeynep Akata
The paper introduces component routing for self‑improving GUI agents, separating experience into locators, procedures, state facts, and lessons, and directing each to either the model weights or the prompt context. Experiments across three backbone families, two environments, and multiple seeds show that routing improves performance over whole‑trajectory baselines, with a rule based on recurrence and state‑conditionality accurately predicting the optimal destination. The study also analyzes how training dynamics and producer‑consumer differences affect the value of each destination, revealing that readout decreases for frequently recurring items when written to weights, while context gains grow with the information gap and weight gains shrink with the policy gap.
By Beining Wu, Zihao Ding, Jun Huang
The paper examines the performance of evolutionary transfer learning and TD(lambda) in the three‑dimensional chess game Dragonchess. By re‑implementing the engine in C++ to accelerate play, the authors ran 10,000 games with statistical confidence, showing both adaptive methods outperform all other agents in a round‑robin tournament. The results indicate no significant performance difference between the evolved and learned evaluation functions, demonstrating the effectiveness of adaptive techniques in complex, novel game domains.
By Jim O'Connor, Annika Hoag, Sarah Goyette, Gary B. Parker
Mingbird is a local‑first agent harness designed for small open‑weight language models (2–9 B) that run on ordinary laptops. It introduces ten mechanisms—such as a byte‑level net‑zero prefill budget, a finish gate that re‑reads the task before accepting completion, and signature‑level loop detection—to address common failure modes that arise from the harness rather than the model itself. In controlled experiments on the LRAB benchmark and the $ au^2$‑bench, Mingbird achieves higher overall scores (0.886 and 0.856 respectively) compared to other harnesses, and its ablation studies show that each mechanism contributes measurable performance gains.
By Hao Wang, Ting Huang
ScholarCatalyst is a new benchmark that evaluates how well AI systems can retrieve research papers that inspire new work. The dataset was created by having 184 lead authors of 207 recent computer science papers annotate which earlier papers helped their projects, providing detailed rationales. The benchmark tests retrieval from the literature available at the start of a project, revealing that current agentic search and even advanced models like Claude Fable 5.1 perform only modestly better than simple embedding retrieval.
By Sohyeon Kim, Yoonho Lee, Bo Liu, Dayoon Ko, Rulin Shao, Seungone Kim, Graham Neubig, Pang Wei Koh, Aakanksha Chowdhery, Akari Asai, Omar Khattab, Yejin Choi, Gunhee Kim, Chelsea Finn
The paper investigates how inherited state affects sub-agent performance in multi-agent frameworks, comparing three inheritance policies—Reset, Selective, and Full—across a ladder of Qwen3 models. It finds that reliance on stale state decreases with model capability, but a mid-capability model (Qwen3‑1.7B) exhibits a statistically significant local minimum of net harm, defining a "danger band." Selective handoff consistently improves accuracy over Full, especially within the danger band, while a fixed-threshold router fails on other datasets.
By Jundong Hu, Shekar Ramachandran
The paper introduces a method for generating legible plans in arbitrary PDDL domains by extending prior legibility research to classical planning without custom planners. It incorporates a second‑order theory of mind to estimate the observer’s perspective, enabling robots to implicitly communicate goals in human‑robot teaming. Benchmark results show that increasing legibility typically trades off with plan efficiency, and a regularizing factor is needed to balance the two.
By Michele Persiani, Thomas Hellstr\"om
The paper introduces Nous, a framework that separates learning, calibration, and revision certification for belief‑based agent memory. It shows that learning and certifying useful memory decisions can require quadratically fewer records than source calibration, and provides finite‑sample certificates for policy improvement without needing to recover source reliability. Experiments on MiniGrid environments demonstrate that the new certificate reliably accepts improvements over incumbents, outperforming earlier methods.
By Pranav Singh
arXiv:2610.00188v1 Announce Type: cross
Abstract: Voxel-based volumetric mapping is fundamental to 3D reconstruction, yet fixed-resolution grids remain inherently inefficient - wasting memory in unif...
By Alpay Ozkan, Tunc Ozan Aydin, Marc Pollefeys, Jelena Trisovic, Daniel Barath
arXiv:2610.00388v1 Announce Type: cross
Abstract: Reinforcement learning enables large language model (LLM) agents to learn multi-step behaviors through interaction with their environments. However,...
By Bo-Wen Zhang, Junwei He, Maoqi Liu, Feiran Li, Song-Lin Lv, Wentao Ma, Rongyi Lin, Shuhan Zhong, Lan-Zhe Guo
arXiv:2610.00400v1 Announce Type: cross
Abstract: Multi-turn attacks on agentic systems can compose individually permissible actions into harmful outcomes, challenging defenses that assess actions or...
By Haoyu Wang, Wei Zhao, Yedi Zhang, Christopher M. Poskitt, Jun Sun
arXiv:2610.00592v1 Announce Type: cross
Abstract: In partially observable reinforcement learning (RL), a later observation can make stored information obsolete or change what it implies for the next...
By Oleg Shchendrigin, Egor Cherepanov, Aleksandr I. Panov, Alexey K. Kovalev
arXiv:2610.00675v1 Announce Type: cross
Abstract: Evolutionary approaches to LLM-driven discovery often generate new programs from a small set of selected ancestors. This keeps contexts manageable bu...
By Bo Yuan, Wenqian Ye, Zelin Zhao, Lama Moukheiber, Henry Kautz, Aidong Zhang, Yongxin Chen
arXiv:2610.00758v1 Announce Type: cross
Abstract: By learning transferable rewards, inverse reinforcement learning (IRL) enables counterfactual evaluation of agents under modified environments. Such...
By Allen Tran, Jia Wan, Nathan Kallus, Aur\'elien Bibaut
arXiv:2610.00838v1 Announce Type: cross
Abstract: Agentic reinforcement learning (RL) trains a large language model (LLM) to act over long, multi-step interactions. However, a single localized error...
By Xinchen Du, Zhengze Zhou, Wenhui Zhu, Han Yu, Sen Na, Rohit Jain, Alborz Geramifard
arXiv:2610.00948v1 Announce Type: cross
Abstract: The executable harness surrounding a GUI model determines how observations are assembled, actions are executed, and verification, recovery, and termi...
By Geyi Yang, Zikun Qu, Xiang Li, Zhiyong Wang, Min Zhang, Shipei Zeng, Zhongxiang Dai
arXiv:2610.01223v1 Announce Type: cross
Abstract: Time-series anomaly detection trades off predictive accuracy, computational efficiency, and interpretability. We use a large language model not as th...
By David Berghaus