arXiv subjects

Kaiyuan Chen

Publications and source records attributed to Kaiyuan Chen.

At least 19 recordsLinked to original sources

arXiv2026

Recursive Experiential-Working Memory Evolution for Long-Horizon Agent Harnesses

Recursive self-improvement (RSI) remains hard in long-horizon tasks, where growing histories obscure the task state and misalign skill invocation. We introduce Recuris, a recursive Experiential-Working Memory architecture for long-horizon agent harnesses, in which Working Memory tracks task progress and guides skill selection from Experiential Memory, grounding skill use in current needs rather than the full history. This coupling also turns execution into structured evidence that localizes failures to specific memory components. Across tasks, a fixed Meta-Agent turns that evidence into localized, validation-gated updates to Skill Memory that reshape execution and yield new evidence, forming a bounded recursive memory-evolution loop. Across four long-horizon benchmarks and ten models, Recuris improves task success in 35 of the 37 completed model-benchmark pairs, carrying frontier models to SOTA-level task success: on tau-bench it adds +17.8 points to GPT-5.6 Sol and +15.6 to Claude Opus 5, taking Opus 5 to 87.9%, and +16.6/+13.5 points on Qwen3.6-27B/35B on SkillFlow. The advantage widens as the interaction horizon grows, to +32.2 points on the longest tasks, and common long-horizon failures fall by up to 80%. These results position recursively evolving memory as a scalable foundation for RSI, enabling agents to continuously transform accumulated experience into increasingly effective long-horizon behavior. Code: https://github.com/Gen-Verse/Recuris

cs.AI↗

arXiv2026

Revisiting the "Push-T" Robot Manipulation Task with Agentic Robotics

Push-T is an iconic benchmark for learning manipulation policies from human demonstrations. The robot must use a single point of contact to push a T-shaped block into a target pose. In this short paper, we revisit the Push-T task in the context of emerging advances in Agentic Robotics where an LLM coding agent -- Claude Code with Fable 5 -- is prompted to create an algorithmic solution that does not require any demonstration data. We study how effective the agentic coding loop can solve the Push-T task, and compare the resulting code as policy with the visuomotor imitation learning policy. Results suggest that the agent found the 2D gym simulation online, and used sim experiments to learn push mechanics, iteratively optimizing to achieve 100% success rate using 46% fewer steps than the best diffusion policy trained with 200 human demonstrations. The coding agent also solve extensions from T to the full alphabet (Push-A to Push-Z) using a self generated curriculum and generated simulation code for the Franka and UR5 robot arms in 3D cross-embodiment simulations with visual feedback. Videos, policies and details will be posted online.

cs.RO↗

arXiv2026

BabyVision: Visual Reasoning Beyond Language

While humans develop core visual skills long before acquiring language, contemporary Multimodal LLMs (MLLMs) still rely heavily on linguistic priors to compensate for their fragile visual understanding. We uncovered a crucial fact: state-of-the-art MLLMs consistently fail on basic visual tasks that humans, even 3-year-olds, can solve effortlessly. To systematically investigate this gap, we introduce BabyVision, a benchmark designed to assess core visual abilities independent of linguistic knowledge for MLLMs. BabyVision spans a wide range of tasks, with 388 items divided into 22 subclasses across four key categories. Empirical results and human evaluation reveal that leading MLLMs perform significantly below human baselines. Gemini3-Pro-Preview scores 49.7, lagging behind 6-year-old humans and falling well behind the average adult score of 94.1. These results show despite excelling in knowledge-heavy evaluations, current MLLMs still lack fundamental visual primitives. Progress in BabyVision represents a step toward human-level visual perception and reasoning capabilities. We also explore solving visual reasoning with generation models by proposing BabyVision-Gen and automatic evaluation toolkit. Our code and benchmark data are released at https://github.com/UniPat-AI/BabyVision for reproduction.

cs.CV↗

arXiv2026

RoboVista: Evaluating Vision Language Models for Diverse Robot Applications

Diverse applications for robotics, such as industry and agriculture, require robots to operate across various embodiments, changing visual conditions, and complex planning. Vision-Language Models (VLMs) offer a promising foundation for general-purpose and interpretable robotic reasoning. Aligning VLMs with diverse robot applications requires a modular understanding of the individual decision components that underlie robotic behavior. Capturing such structure is challenging for conventional robot benchmarks that are primarily based on teleoperated, end-to-end datasets. We propose Robot Question Answering (RQA), a modular evaluation framework and RoboVista, a benchmark curated from real robotic systems, research papers, and expert annotations. RoboVista contains 474 Visual Question Answering (VQA) instances with human annotated reasoning and covers 39 unique task types in agricultural, industrial, domestic, surgical robotics, autonomous driving, and open robot datasets. Experiments on RoboVista show that state-of-the-art VLMs exhibit substantial gaps. Physical robot experiments suggest strong correlation between RoboVista performance and real-world task execution.

cs.RO↗

arXiv2026

GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks

For robots to work reliably in commercial and industrial applications, can recent advances in agentic coding systems combine interpretable robot programming with the open-world adaptability of model-free policies? We focus on "Variational Automation" (VA), a class of tasks that have larger variations in object geometry and pose than fixed automation. Model-free policies often struggle to close the reliability gap for VA tasks, which must be executed persistently and reliably in commercial and industrial applications. Motivated by prior work on Task and Motion Planning (TAMP) and the Robot Operating System (ROS), we introduce Graph-as-Policy (GaP), a multi-agent coding harness that generates directed computation graphs with perception, planning, and control nodes from a Modular Open Robot Skill Library (MORSL). GaP then generates an internal simulation environment to rehearse task instances with different graphs in parallel to iteratively refine the graph structure and parameters to improve success rates and throughput. Evaluation with 8 new open VA task benchmarks, 4 in-simulation and 4 in real-world, suggests that GaP can achieve success rates that significantly outperform baselines. Details, code, and data can be found online: https://graph-robots.github.io/gap

cs.RO↗

arXiv2026

Optimization-based Safe Trajectory Planning for Autonomous Ground Vehicle in Multi-Floor Scenarios

The development of trajectory planning strategies for autonomous ground vehicles (AGVs) represents a prevailing research interest within the domain of intelligent transportation systems. This paper introduces a trajectory planning framework tailored for multi-floor scenarios. The framework consists of two main modules: the task planning module and the trajectory planning module. The task planning module involves a strategic selection phase, where a task planning strategy based on generalized voronoi diagrams (GVD) and multi-objective algorithms is proposed to select the floor exits for each floor. The trajectory planning module utilizes optimization-based methods to generate high-quality trajectories, and a warm-started hierarchical planning framework is designed to ensure rapid convergence. Additionally, for handling complex obstacle constraints, a correlation constraint calculation method is designed for reducing obstacle constraints in trajectory planning. Finally, the feasibility and effectiveness of the proposed framework are verified through simulations.

cs.RO↗

arXiv2026

Unveiling Downstream Performance Scaling of LLMs: A Clustering-Based Perspective

The escalating scale and cost of Large Language Models (LLMs) training necessitate accurate pre-training prediction of downstream task performance for comprehensive understanding of scaling properties. This is challenged by: 1) the emergence phenomenon, where unpredictable capabilities appearing suddenly at critical model scales; and 2) uneven task difficulty and inconsistent performance scaling patterns, leading to high metric variability. Current prediction methods lack accuracy and reliability. We propose a Clustering-On-Difficulty (COD) framework for downstream performance prediction. The COD framework clusters tasks by their difficulty scaling features, thereby constructing a more stable and predictable task subset that exhibits well-behaved scaling characteristics with the increase of compute budget. We adopt a performance scaling law to predict cluster-wise performance with theoretical support. Predictable subset performance acts as an intermediate predictor for the full evaluation set. We further derive a mapping function to accurately extrapolate the performance of the subset to the full set. Applied to an LLM with 70B parameters, COD achieved a 1.55\% average prediction error across eight key LLM benchmarks, thus providing actionable insights for scaling properties and training monitoring during LLM pre-training.

cs.CL↗

arXiv2026

\$OneMillion-Bench: How Far are Language Agents from Human Experts?

As language models (LMs) evolve from chat assistants to long-horizon agents capable of multi-step reasoning and tool use, existing benchmarks remain largely confined to structured or exam-style tasks that fall short of real-world professional demands. To this end, we introduce \$OneMillion-Bench \$OneMillion-Bench, a benchmark of 400 expert-curated tasks spanning Law, Finance, Industry, Healthcare, and Natural Science, built to evaluate agents across economically consequential scenarios. Unlike prior work, the benchmark requires retrieving authoritative sources, resolving conflicting evidence, applying domain-specific rules, and making constraint decisions, where correctness depends as much on the reasoning process as the final answer. We adopt a rubric-based evaluation protocol scoring factual accuracy, logical coherence, practical feasibility, and professional compliance, focused on expert-level problems to ensure meaningful differentiation across agents. Together, \$OneMillion-Bench provides a unified testbed for assessing agentic reliability, professional depth, and practical readiness in domain-intensive scenarios.

cs.LG↗

arXiv2026

SAPO: Self-Adaptive Process Optimization Makes Small Reasoners Stronger

Existing self-evolution methods overlook the influence of fine-grained reasoning steps, which leads to the reasoner-verifier gap. The computational inefficiency of Monte Carlo (MC) process supervision further exacerbates the difficulty in mitigating the gap. Motivated by the Error-Related Negativity (ERN), which the reasoner can localize error following incorrect decisions, guiding rapid adjustments, we propose a Self-Adaptive Process Optimization (SAPO) method for self-improvement in Small Language Models (SLMs). SAPO adaptively and efficiently introduces process supervision signals by actively minimizing the reasoner-verifier gap rather than relying on inefficient MC estimations. Extensive experiments demonstrate that the proposed method outperforms most existing self-evolution methods on two challenging task types: mathematics and code. Additionally, to further investigate SAPO's impact on verifier performance, this work introduces two new benchmarks for process reward models in both mathematical and coding tasks.

cs.CL↗

arXiv2026

AgentIF-OneDay: A Task-level Instruction-Following Benchmark for General AI Agents in Daily Scenarios

The capacity of AI agents to effectively handle tasks of increasing duration and complexity continues to grow, demonstrating exceptional performance in coding, deep research, and complex problem-solving evaluations. However, in daily scenarios, the perception of these advanced AI capabilities among general users remains limited. We argue that current evaluations prioritize increasing task difficulty without sufficiently addressing the diversity of agentic tasks necessary to cover the daily work, life, and learning activities of a broad demographic. To address this, we propose AgentIF-OneDay, aimed at determining whether general users can utilize natural language instructions and AI agents to complete a diverse array of daily tasks. These tasks require not only solving problems through dialogue but also understanding various attachment types and delivering tangible file-based results. The benchmark is structured around three user-centric categories: Open Workflow Execution, which assesses adherence to explicit and complex workflows; Latent Instruction, which requires agents to infer implicit instructions from attachments; and Iterative Refinement, which involves modifying or expanding upon ongoing work. We employ instance-level rubrics and a refined evaluation pipeline that aligns LLM-based verification with human judgment, achieving an 80.1% agreement rate using Gemini-3-Pro. AgentIF-OneDay comprises 104 tasks covering 767 scoring points. We benchmarked four leading general AI agents and found that agent products built based on APIs and ChatGPT agents based on agent RL remain in the first tier simultaneously. Leading LLM APIs and open-source models have internalized agentic capabilities, enabling AI application teams to develop cutting-edge Agent products.

cs.CL↗

arXiv2025

Discovery of two little red dots transitioning into quasars

James Webb Space Telescope (JWST) has uncovered a new population of compact objects that show a unique V-shaped spectral energy distribution (SED) in the UV and optical wavelength range. These so-called "little red dots" (LRDs) often exhibit broad Balmer emission lines, indicative of the presence of active galactic nuclei (AGNs). They generally lack detection of X-ray, radio, and mid-IR radiation, which is fundamentally different from typical AGNs. Various models, including super-Eddington-accreting black holes enshrouded in dense and dust-poor gas, have been proposed to explain these features. However, the nature of LRDs remains debated, and their evolutionary fate is unclear. Here we report two unusual LRDs at redshift z = 2.868 and 2.925 that are in a transitional phase towards typical AGNs. Their V-shaped SEDs, compact optical morphology, and broad emission lines satisfy the defining criteria of LRDs. On the other hand, they exhibit intense X-ray, radio, and mid-IR radiation that is much stronger than previously known LRDs. These hybrid properties suggest that the dense gas envelope around their central black holes is dispersing, allowing high-energy photons and radio emission to escape. Meanwhile, a dust torus is forming. This finding provides a direct insight into the nature of LRDs and indicates that at least some LRDs will evolve into AGNs or quasars at later times.

astro-ph.GA↗

arXiv2025

Virtual Width Networks

We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN decouples representational width from backbone width, expanding the embedding space while keeping backbone compute nearly constant. In our large-scale experiment, an 8-times expansion accelerates optimization by over 2 times for next-token and 3 times for next-2-token prediction. The advantage amplifies over training as both the loss gap grows and the convergence-speedup ratio increases, showing that VWN is not only token-efficient but also increasingly effective with scale. Moreover, we identify an approximately log-linear scaling relation between virtual width and loss reduction, offering an initial empirical basis and motivation for exploring virtual-width scaling as a new dimension of large-model efficiency.

cs.LG↗

arXiv2025

LAVQA: A Latency-Aware Visual Question Answering Framework for Shared Autonomy in Self-Driving Vehicles

When uncertainty is high, self-driving vehicles may halt for safety and benefit from the access to remote human operators who can provide high-level guidance. This paradigm, known as {shared autonomy}, enables autonomous vehicle and remote human operators to jointly formulate appropriate responses. To address critical decision timing with variable latency due to wireless network delays and human response time, we present LAVQA, a latency-aware shared autonomy framework that integrates Visual Question Answering (VQA) and spatiotemporal risk visualization. LAVQA augments visual queries with Latency-Induced COllision Map (LICOM), a dynamically evolving map that represents both temporal latency and spatial uncertainty. It enables remote operator to observe as the vehicle safety regions vary over time in the presence of dynamic obstacles and delayed responses. Closed-loop simulations in CARLA, the de-facto standard for autonomous vehicle simulator, suggest that that LAVQA can reduce collision rates by over 8x compared to latency-agnostic baselines.

cs.RO↗

arXiv2025

Robust Set Partitioning Strategy for Malicious Information Detection in Large-Scale Internet of Things

With the rapid development of the Internet of Things (IoT), the risks of data tampering and malicious information injection have intensified, making efficient threat detection in large-scale distributed sensor networks a pressing challenge. To address the decline in malicious information detection efficiency as network scale expands, this paper investigates a robust set partitioning strategy and, on this basis, develops a distributed attack detection framework with theoretical guarantees. Specifically, we introduce a gain mutual influence metric to characterize the inter-subset interference arising during gain updates, thereby revealing the fundamental reason for the performance gap between distributed and centralized algorithms. Building on this insight, the set partitioning strategy based on Grassmann distance is proposed, which significantly reduces the computational cost of gain updates while maintaining detection performance, and ensures that the distributed setting under subset partitioning preserves the same theoretical performance bound as the baseline algorithm. Unlike conventional clustering methods, the proposed set partitioning strategy leverages the intrinsic observational features of sensors for robust partitioning, thereby enhancing resilience to noise and interference. Simulation results demonstrate that the proposed method limits the performance gap between distributed and centralized detection to no more than 1.648$\%$, while the computational cost decreases at an order of $O(1/m)$ with the number of subsets $m$. Therefore, the proposed algorithm effectively reduces computational overhead while preserving detection accuracy, offering a practical low-cost and highly reliable security detection solution for edge nodes in large-scale IoT systems.

cs.DC↗

arXiv2025

Enhanced Trust Region Sequential Convex Optimization for Multi-Drone Thermal Screening Trajectory Planning in Urban Environments

The rapid detection of abnormal body temperatures in urban populations is essential for managing public health risks, especially during outbreaks of infectious diseases. Multi-drone thermal screening systems offer promising solutions for fast, large-scale, and non-intrusive human temperature monitoring. However, trajectory planning for multiple drones in complex urban environments poses significant challenges, including collision avoidance, coverage efficiency, and constrained flight environments. In this study, we propose an enhanced trust region sequential convex optimization (TR-SCO) algorithm for optimal trajectory planning of multiple drones performing thermal screening tasks. Our improved algorithm integrates a refined convex optimization formulation within a trust region framework, effectively balancing trajectory smoothness, obstacle avoidance, altitude constraints, and maximum screening coverage. Simulation results demonstrate that our approach significantly improves trajectory optimality and computational efficiency compared to conventional convex optimization methods. This research provides critical insights and practical contributions toward deploying efficient multi-drone systems for real-time thermal screening in urban areas. For reader who are interested in our research, we release our source code at https://github.com/Cherry0302/Enhanced-TR-SCO.

cs.RO↗

arXiv2025

UAV-UGV Cooperative Trajectory Optimization and Task Allocation for Medical Rescue Tasks in Post-Disaster Environments

In post-disaster scenarios, rapid and efficient delivery of medical resources is critical and challenging due to severe damage to infrastructure. To provide an optimized solution, we propose a cooperative trajectory optimization and task allocation framework leveraging unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). This study integrates a Genetic Algorithm (GA) for efficient task allocation among multiple UAVs and UGVs, and employs an informed-RRT* (Rapidly-exploring Random Tree Star) algorithm for collision-free trajectory generation. Further optimization of task sequencing and path efficiency is conducted using Covariance Matrix Adaptation Evolution Strategy (CMA-ES). Simulation experiments conducted in a realistic post-disaster environment demonstrate that our proposed approach significantly improves the overall efficiency of medical rescue operations compared to traditional strategies. Specifically, our method reduces the total mission completion time to 26.7 minutes for a 15-task scenario, outperforming K-Means clustering and random allocation by over 73%. Furthermore, the framework achieves a substantial 15.1% reduction in total traveled distance after CMA-ES optimization. The cooperative utilization of UAVs and UGVs effectively balances their complementary advantages, highlighting the system's scalability and practicality for real-world deployment.

cs.RO↗

arXiv2025

Trajectory Optimization for UAV-Based Medical Delivery with Temporal Logic Constraints and Convex Feasible Set Collision Avoidance

This paper addresses the problem of trajectory optimization for unmanned aerial vehicles (UAVs) performing time-sensitive medical deliveries in urban environments. Specifically, we consider a single UAV with 3 degree-of-freedom dynamics tasked with delivering blood packages to multiple hospitals, each with a predefined time window and priority. Mission objectives are encoded using Signal Temporal Logic (STL), enabling the formal specification of spatial-temporal constraints. To ensure safety, city buildings are modeled as 3D convex obstacles, and obstacle avoidance is handled through a Convex Feasible Set (CFS) method. The entire planning problem-combining UAV dynamics, STL satisfaction, and collision avoidance-is formulated as a convex optimization problem that ensures tractability and can be solved efficiently using standard convex programming techniques. Simulation results demonstrate that the proposed method generates dynamically feasible, collision-free trajectories that satisfy temporal mission goals, providing a scalable and reliable approach for autonomous UAV-based medical logistics.

eess.SY↗

arXiv2025

Robo2VLM: Visual Question Answering from Large-Scale In-the-Wild Robot Manipulation Datasets

Vision-Language Models (VLMs) acquire real-world knowledge and general reasoning ability through Internet-scale image-text corpora. They can augment robotic systems with scene understanding and task planning, and assist visuomotor policies that are trained on robot trajectory data. We explore the reverse paradigm - using rich, real, multi-modal robot trajectory data to enhance and evaluate VLMs. In this paper, we present Robo2VLM, a Visual Question Answering (VQA) dataset generation framework for VLMs. Given a human tele-operated robot trajectory, Robo2VLM derives ground-truth from non-visual and non-descriptive sensory modalities, such as end-effector pose, gripper aperture, and force sensing. Based on these modalities, it segments the robot trajectory into a sequence of manipulation phases. At each phase, Robo2VLM uses scene and interaction understanding to identify 3D properties of the robot, task goal, and the target object. The properties are used to generate representative VQA queries - images with textural multiple-choice questions - based on spatial, goal-conditioned, and interaction reasoning question templates. We curate Robo2VLM-1, a large-scale in-the-wild dataset with 684,710 questions covering 463 distinct scenes and 3,396 robotic manipulation tasks from 176k real robot trajectories. Results suggest that Robo2VLM-1 can benchmark and improve VLM capabilities in spatial and interaction reasoning.

cs.RO↗