Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
arXiv:2607.05205v2 Announce Type: replace Abstract: Fast and reliable motion detection is essential for machine vision and autonomous systems operating in dynamic environments. This work integrates e...
ComplicitSplat is a novel black‑box attack that leverages 3D Gaussian Splatting (3DGS) shading to create viewpoint‑specific camouflage, embedding adversarial content into scene objects that is only visible from certain angles. The method does not require access to model architecture or weights and can successfully fool a range of popular object detectors—including single‑stage, multi‑stage, and transformer‑based models—on both real‑world physical objects and synthetic scenes. This demonstrates that downstream models using 3DGS are vulnerable to adversarial manipulation.
PV-WM is a history‑only world model that jointly predicts pedestrian root motion, 15‑joint articulation, and vehicle kinematic states in a synchronized heterogeneous state. It uses recurrent updates to generate pedestrian and vehicle motion chunks, reconstructing vehicle boxes from predicted center, heading, and observed extent, and recomputes pedestrian‑vehicle geometry after each transition. Compared to a one‑shot predictor, PV‑WM reduces Root ADE by 12.7% and MPJPE by 14.8%, and across 824 Waymo contexts it lowers Root ADE by 5.2%, MPJPE by 7.6%, P‑V distance error by 11.9%, and oriented‑box closest‑approach error by 5.8%, while using 57.1% fewer parameters, 96.5% fewer FLOPs, and 25.5% lower p95 latency.
StochBench is a new Lean 4 benchmark comprising 450 graduate‑level stochastic‑process problems, each linked to its natural‑language source. It covers a wide range of topics—finite and countable Markov chains, renewal processes, random walks, martingales, stopping times, queues, Brownian motion, stochastic calculus, weak convergence, and Poisson and continuous‑time Markov processes—addressing a field underrepresented in Mathlib. An Opus 4.8‑based agent achieved a 34.9% proof rate (157/450) within a 15‑minute per‑problem limit, demonstrating the benchmark’s difficulty for advanced provers.
The paper introduces a multi‑modal late‑fusion perception pipeline for object detection and tracking in autonomous racing. It combines independent detections from cameras, LiDARs, and RADARs to produce timely and robust state estimates of surrounding vehicles. The tracking framework compensates for detection delays and incorporates vehicle dynamics and track layout knowledge, and its effectiveness is confirmed through real‑world experiments in diverse critical scenarios.
PlannerForge is a unified LLM‑agent framework that covers the entire scenario‑based testing pipeline for autonomous driving systems, from scenario generation to ADS assessment, and adds ADS enhancement and benchmarking stages. It was evaluated with ten off‑the‑shelf LLMs across all tasks and five prompt conditions, achieving best‑per‑task scores between 0.88 and 1.00 and matching commercial APIs with open‑source models such as Qwen3.6:35B. The end‑to‑end chaining retains 83% of seed queries for commercial backends and 78% for open‑source, outperforming existing tools like Scenario Factory 2.0 and BM25 in natural‑language generation, attribute realization, and physically valid edits. whyItMatters":"PlannerForge demonstrates that a single LLM‑based system can streamline and improve the fragmented scenario‑based testing workflow for autonomous driving, achieving high performance without domain‑specific fine‑tuning."
The paper introduces DroidTool, a framework that enables Android GUI agents to automatically generate and integrate tool actions—Python functions that manipulate application states—into their hybrid action space. DroidTool follows an agentic workflow with stages of proposal, implementation, test generation, execution, and repair, and it constructs relational tests to ensure comprehensive verification. Experiments on benchmarks such as AndroidWorld, B-MoCA, and MobileSafetyBench show that agents using DroidTool’s self-generated tools outperform GUI-only agents by about 4.47% in performance while reducing interactions by roughly 20.05%.
The paper "Inferring the Unspoken: Aligning Embodied Agents with Implicit Preferences" addresses the challenge of natural-language instructions that omit details needed for embodied action. It introduces the Preference-based Planning (PbP) benchmark, comprising 5,000 evaluation groups and 290 preferences across three levels, to systematically evaluate agents’ ability to infer latent user preferences from a few demonstrations. The authors propose the two-stage Inferring the Unspoken (InTU) framework, which first verbalizes inferred preferences from multimodal demonstrations and then generates action plans conditioned on that explicit representation, showing that explicit verbalization improves alignment and robustness compared to direct end-to-end planning.
The paper introduces MIRAGE, a benchmark of 750 multi‑step decision tasks designed to test autonomous web agents’ investigative abilities across Wikipedia Forensics, Shopping Admin adjudication, and Reddit Moderation. Each task contains a misleading visible context and a hidden context that holds decisive evidence, allowing performance to be broken down into Investigation, Reasoning, and Decision Accuracy, with an added Investigative Hallucination Rate. Evaluation of eight LLM agents reveals three consistent patterns: agents often reach relevant pages but fail to extract decisive evidence, procedural hints improve investigation but not decision accuracy on Wikipedia tasks, and 12.6% of trajectories include fabricated facts.
The paper investigates how to apply Reinforcement Learning with Verifiable Rewards (RLVR) to large language models across multiple domains. It compares two training paradigms—mixed multi-task RLVR and separate RLVR followed by model merging—using tasks such as math, coding, science, instruction following, and agent. Experiments show that RLVR across domains causes minimal interference and that reasoning-intensive domains can synergize, with insights drawn from information constraints, prediction behavior, and self-verification.
AutoFyn is a new agent framework based on the Expert Iteration algorithm that updates a frozen model’s persistent state using verified reward signals instead of changing model weights. Each iteration starts with a fresh model session, and only durable information—such as memory files, reports, and repository state—is carried over through explicit interfaces. The system orchestrates exploration, planning, and specialized agents, while a task‑grounded verifier supplies objective rewards that are distilled back into the persistent state to guide the next round. AutoFyn has been applied to olympiad mathematics, data science, and cybersecurity, outperforming baseline models on the 2026 International Mathematical Olympiad, topping the Spider 2.0 dbt benchmark, and generating sixteen maintainer‑confirmed vulnerability advisories across several popular software projects.
SCAFFOLD is a self‑improving framework for visual web agents that automatically induces parametric, executable skills from successful trajectories and organizes them into a recursively composed hierarchy. It compresses the skill library using a minimum‑description‑length criterion and behavioral equivalence checks, and periodically distills these skills back into model weights to internalize the abstractions. Experiments on WebArena, VisualWebArena, and Online‑Mind2Web show that SCAFFOLD raises success rates by 11.1–17.2 absolute points over the best skill‑augmented baseline and continues to improve across five self‑improvement iterations without collapsing the library.
The study introduces a World Values Survey–grounded simulation framework to test whether large language model agents can faithfully represent diverse human value systems. In about 4,000 conversations with 1,200 personas across three models, more than half of the agents failed to express their assigned value profiles from the start, and only 2–7% drifted over time. The results show systematic deviations from the intended value distributions and reveal that simulated dialogues differ from human discussions in their balance of stylistic consistency and semantic diversity.
The paper presents a deep active inference framework for real‑world robotic navigation that combines a diffusion policy with a multiple‑timescale recurrent state‑space model. The diffusion policy generates diverse candidate actions, while the state‑space model predicts long‑horizon outcomes, allowing the system to select actions that minimize expected free energy. Experiments show higher success rates and fewer collisions, especially in exploration‑heavy scenarios, demonstrating the effectiveness of this unified exploration and goal‑directed approach.
EdgeMem is a new agent-memory method that preserves original interaction turns and organizes them using complementary content, temporal, and episodic cues via a multi‑anchor hypergraph. It performs lightweight local processing, returning source evidence directly and reserving LLM use only for final answer generation. Experiments on LoCoMo and LongMemEval‑S demonstrate strong retrieval and memory‑grounded question answering, with EdgeMem achieving the highest strict‑judge score among seven systems on LoCoMo while requiring no generative‑LLM calls for construction and retrieval.
The paper proposes a framework for improving technical‑service large language model agents by integrating latent logic augmentation, robust noise reduction, and hybrid reward modeling. It enhances supervised fine‑tuning data with Planning‑Aware Trajectory Modeling and Reasoning Augmentation, creates dual‑filtered multiple ground truths, and trains the policy using a hybrid reward that combines a reranker with an LLM‑as‑a‑Judge. Applied to real Cloud technical‑service tasks, the adapted Qwen3‑4B model achieves the highest Multi‑ECS score, reduces reward cost, and attains the highest production adoption rate.
IdeaAMBIG is a benchmark that evaluates how well research‑idea specifications are ready for implementation. It contains 660 evidence‑grounded instances, including 163 real‑world gaps from reproducibility reports and GitHub issues and 497 synthetic gaps added to codification‑ready references. The benchmark tests three capabilities—codification‑readiness assessment, defect localization, and clarification action generation—across 13 large language models, revealing that defect localization is the main bottleneck while clarification given the defect performs much better.
The paper investigates whether modifying a model’s internal activations—through activation steering—actually changes the computational state used in subsequent processing or merely biases the computation toward a desired output. In a controlled state‑tracking task, editing a trace‑supervised register causes the model to apply the next operation to the edited state, confirming that the edit changes the state. However, in two large‑language‑model settings (Qwen and Llama), mean activation steering does not replicate the natural internal configuration used during task execution; steering vectors are far larger than typical natural changes and achieve only a fraction of the effect of full‑layer patching. Thus, a successful steering intervention does not necessarily reproduce the natural target activation at the intervention layer, and should be interpreted as a state change only when later computation actually uses the edited value in the intended semantic way.
The paper introduces Many-Tier Instruction Hierarchy (ManyIH), a new framework for resolving conflicts among instructions with arbitrarily many privilege levels in large language model agents. It presents ManyIH-Bench, a benchmark featuring 853 agentic tasks that require navigating up to 12 levels of conflicting instructions across 46 real-world agents. Experiments show current models achieve only about 40% accuracy when instruction conflict scales, highlighting a gap in fine-grained, scalable conflict resolution.
The article titled "The convergent laboratory: when AI reasoning, autonomous experiments, high performance and quantum computing reshape chemistry" discusses insights from the TPC26 conference, where leaders from academia, national laboratories, and industry examined how AI, autonomous agents, self-driving labs, high‑performance computing, and quantum computing converge to accelerate materials science discovery. It presents firsthand experiences from researchers at the forefront of these technologies and argues that their simultaneous maturation marks a tipping point for transformative advances and productive disruption in chemical sciences.