Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
Marketers now deploy generative AI agents as synthetic consumers to pretest visual assets such as logos, packaging, and advertising at a fraction of human-panel cost. However, this procedure assumes t...
The paper introduces SCOPE, a method that post‑trains computer‑use agents to balance task completion with safety by conditioning actions on environmental risk. It combines supervised fine‑tuning on three trajectory types—capability demonstrations, safe continuations, and explicit refusals—followed by reinforcement learning to improve performance. Experiments starting from Qwen3.5‑9B show that SCOPE‑RL achieves high task success and attack‑avoidance rates, outperforming other agents on OSWorld and OS‑BLIND benchmarks.
The paper introduces Chart‑RVR, a reinforcement‑learning framework that trains chart‑reasoning agents to produce monitorable, verifiable outputs. It decomposes reasoning into three auditable stages—Structure, Evidence, and Derivation—allowing stakeholders to trace how the model reads the chart, extracts data, and computes the answer. Experiments on six benchmarks show that Chart‑RVR matches or exceeds state‑of‑the‑art accuracy while delivering rationales that are far more verifiable and evidence‑grounded than existing methods.
A$^2$Safe is a framework for safe and effective Visual Question Answering that aligns counterfactual evidence with adaptive agent collaboration. It uses a Grounded Safety Evidence Board to make safety decisions explicit, enforcing invariance to safety‑irrelevant changes while allowing appropriate transitions when risk‑critical evidence changes. The system achieves a 95.72 SIUO safety score, reduces benign refusals on MOSSBench to 14.67%, and maintains a 78.34 average VQA score with 27.8% token overhead.
The paper studies classifier‑free guidance (CFG) in Flow Matching, showing that strong guidance can distort the generated distribution by shifting the mean and concentrating trajectories. By interpreting Flow Matching as a time‑varying gradient flow, the authors explain how CFG reshapes the underlying potential and propose a training‑free method, Posterior‑Mean‑Capped CFG (PMC‑CFG), that adaptively limits guidance to the strongest feasible level. Experiments on synthetic and large‑scale image‑generation tasks demonstrate that PMC‑CFG reduces distortion and concentration while improving the alignment–diversity trade‑off, especially when nominal guidance is large.
VideoGen-Agent is a multimodal agent that uses multitask agentic reinforcement learning to coordinate external tools for video generation. It learns to augment, generate, and verify videos through multi‑turn interactions, guided by prompts and intermediate observations. On the new VABench benchmark, the agent improves base text‑to‑video performance by 19.1 points, and further upgrades to generation tools raise the score to 86.1, with human raters favoring the upgraded configuration in 84.3% of comparisons.
The paper introduces the ACE framework, which uses spatially explicit inspection to guide embodied exploration. By combining evidence‑grounded perception with exposure‑informed movement, ACE provides a spatially resolved decision paradigm that improves cue assessment and movement direction. Experiments show ACE boosts navigation task success by 18.0% and exploration efficiency by 10.3% over previous baselines.
The paper introduces Trace, a framework that transforms sparse-reward trajectories into executable walkthroughs by identifying progress anchors, propagating credit, and estimating action prerequisites. Trace compiles noisy trajectories into state‑conditioned, verifiable procedures that remove loops and detours, enabling reuse, intermediate‑state resumption, and programmatic verification. Experiments on J‑TTL, WebShop, and ScienceWorld with three open‑source LLMs show that Trace outperforms eight baselines, improving average AUC and Final‑$3$ by 30.0% and 40.5% while using fewer inference tokens.
The paper presents a hierarchical Bayesian optimization framework that uses Gaussian process meta-modeling to address the challenges of optimizing complex system-of-systems (SoS) architectures. It handles discrete architectural choices, conditional dependencies, and heterogeneous design variables, improving search efficiency and robustness over conventional surrogate-based methods. The approach is demonstrated on an aircraft-based multi-agent system for wildfire suppression, showing its applicability to large, diverse design spaces with limited simulation budgets.
StepKV introduces a step-aware approach to compressing the key-value cache used during large language model inference, treating reasoning steps as primary units of retention rather than individual tokens. By linking cache entries to the steps that generated them and estimating each step’s utility from trajectory signals, StepKV assigns a combined token‑ and step‑level score to guide pruning. Experiments on multi‑hop question answering and long‑horizon web reasoning show that StepKV maintains accuracy even under tight cache budgets, outperforming token‑level baselines that suffer sharp performance drops.
OpenBlock is an open, adaptive tile‑matching platform that uses a dual‑track content‑generation system: a deterministic rule‑based generator and an optional learned generator, both verified to ensure every piece set is fully placeable. A self‑play reinforcement‑learning agent diagnoses that long‑bar pieces become largely unplaceable at high board fill, indicating board‑state degeneration rather than difficulty drives late‑game failure. In live play, the learned track improves day‑1 retention by 1.8 percentage points and session duration by 7% compared to the rule track alone.
Toollery is a training‑free framework that compresses candidate lists for large language model agents, enabling efficient selection from thousands of skills and tools. It generates user‑intent queries from each skill or tool specification, builds a retrieval index, and limits online selection to a compact top‑k set before the LLM makes its final decision. Evaluations on the SkillRouter benchmark, BFCL‑V4, and a proprietary smart‑cockpit dataset show that Toollery improves recall and end‑to‑end selection while keeping selection costs bounded.
RS-Claw-Evolution is an environment-feedback-driven framework designed to enhance lightweight remote sensing agents for long-horizon tasks. It improves agents through three stages—interaction evolution, experience evolution, and decision evolution—using executable code, failure-aware trajectory generation, and reinforcement learning with multi-dimensional rewards. On Earth-Bench, a Qwen3-4B agent trained with this framework reaches 65.9% accuracy, surpassing larger baselines and approaching GPT-5 performance.
The paper introduces a closed‑loop cyber‑physical system for autonomous model lifecycle management in automotive manufacturing, deployed since 2023. It manages paired physics and reinforcement‑learning models, selecting the best candidate through competitive retraining cycles and a Conductor orchestrator that handles plant‑wide inventories and fallback controls. The system incorporates an operator‑trust gate that rejects 23% of policies that deviate from established practice, achieving 28‑45% process stability improvements with no safety incidents.
The paper introduces a counterfactual tool ranking framework that accounts for authority, historical support, and estimation nuances. Using eleven enterprise-inspired tools, synthetic and real-world experiments on the Berkeley Function Calling Leaderboard, the study compares direct regression and doubly robust (DR) methods, finding that DR performs better in shifted environments while direct regression excels in linear settings. The authors also evaluate Qwen2.5 models on held-out tasks, analyze policy differences under missing support, and present a falsifiable evaluation method with publicly available evidence.
arXiv:2609.24144v1 Announce Type: new Abstract: Group-relative reinforcement learning compares rollouts of the same prompt, but independent environment noise can obscure these comparisons. We study p...
arXiv:2609.24289v1 Announce Type: new Abstract: As Large Language Model (LLM) agents are applied in continuously interactive environments, driving the evolution of their own capabilities becomes a co...
arXiv:2609.24629v1 Announce Type: new Abstract: A/B testing requires large sample sizes, long timelines, and significant costs. When auxiliary predictions of experimental outcomes are available from...
arXiv:2609.24862v1 Announce Type: new Abstract: Agentic time series forecasting concerns systems whose underlying mechanisms evolve, making the relative effectiveness of numerical models, reasoning s...
arXiv:2609.24969v1 Announce Type: new Abstract: As agents are deployed with increased autonomy, even extremely rare events along their stochastic output trajectories can occur and prove catastrophic....