arXiv subjects

Xiatao Sun

Publications and source records attributed to Xiatao Sun.

16 recordsLinked to original sources

arXiv2026

Artificial Foveated Perception for Mitigating Shortcut Learning in Robotic Foundation Models

Robotic foundation models still need task-specific fine-tuning before deployment, and the fine-tuned policies often break under modest changes in scene layout, lighting, or nearby distractors. We trace this brittleness to \textit{shortcut learning}: fine-tuning supervises actions but not the visual evidence the policy uses, so the policy can settle on scene-level correlations that predict the demonstrations without causing success. We propose Artificial Foveated Perception (AFP), a lightweight, policy-agnostic module that takes the same vision and language inputs as existing Vision-Language-Action and World Action Model pipelines and predicts task-conditioned masks over the relevant objects, the robot, and other action-critical regions. During fine-tuning the masks serve as an auxiliary grounding signal that aligns the policy's visual attention with task-relevant regions; the policy architecture is unchanged, and at inference the policy runs on the original observation stream with no AFP call in the control loop. In simulation with four robotic foundation models and on a real robot with $\pi_{0.5}$, AFP improves generalization under environmental perturbations, reduces overfitting, and shortens fine-tuning. Ablations over mask quality and grounding-loss design show that these gains come from directing policy learning toward task-relevant visual evidence. Code, data, and videos are available at https://apollo-lab-yale.github.io/26-CoRL-AFP-website/.

cs.RO↗

arXiv2026

Agents' Last Exam

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long horizon, economically valuable, real world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 sub fields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is below 1%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP relevant impact.

cs.AI↗

arXiv2026

Turning Stale Gradients into Stable Gradients: Coherent Coordinate Descent with Implicit Landscape Smoothing for Lightweight Zeroth-Order Optimization

Zeroth-Order (ZO) optimization is pivotal for scenarios where backpropagation is unavailable, such as memory-constrained on-device learning and black-box optimization. However, existing methods face a stark trade-off: they are either sample-inefficient (e.g., standard finite differences) or suffer from high variance due to randomized estimation (e.g., random subspace methods). In this work, we propose Coherent Coordinate Descent (CoCD), a deterministic, sample-efficient, and budget-aware ZO optimizer. Theoretically, we formalize the notion of gradient coherence and demonstrate that CoCD is equivalent to Block Cyclic Coordinate Descent (BCCD) with ``warm starts,'' effectively converting historical (stale) gradients from a liability into a computational asset. This mechanism enables $O(1)$ query complexity per step while maintaining global descent directions. Furthermore, we derive error bounds revealing a counter-intuitive insight: larger finite-difference step sizes can induce an implicit smoothing effect on the optimization landscape by reducing the effective smoothness constant, thereby improving convergence stability. Experiments on MLP, CNN, and ResNet architectures (up to 270k parameters) demonstrate that CoCD significantly outperforms BCCD in terms of sample efficiency and convergence loss/accuracy, and exhibits superior stability over randomized ZO methods. Our results suggest that deterministic, structure-aware updates offer a superior alternative to randomization for lightweight ZO optimization.

cs.LG↗

arXiv2026

RuleSmith: Multi-Agent LLMs for Automated Game Balancing

Game balancing is a longstanding challenge requiring repeated playtesting, expert intuition, and extensive manual tuning. We introduce RuleSmith, the first framework that achieves automated game balancing by leveraging the reasoning capabilities of multi-agent LLMs. It couples a game engine, multi-agent LLMs self-play, and Bayesian optimization operating over a multi-dimensional rule space. As a proof of concept, we instantiate RuleSmith on CivMini, a simplified civilization-style game containing heterogeneous factions, economy systems, production rules, and combat mechanics, all governed by tunable parameters. LLM agents interpret textual rulebooks and game states to generate actions, to conduct fast evaluation of balance metrics such as win-rate disparities. To search the parameter landscape efficiently, we integrate Bayesian optimization with acquisition-based adaptive sampling and discrete projection: promising candidates receive more evaluation games for accurate assessment, while exploratory candidates receive fewer games for efficient exploration. Experiments show that RuleSmith converges to highly balanced configurations and provides interpretable rule adjustments that can be directly applied to downstream game systems. Our results illustrate that LLM simulation can serve as a powerful surrogate for automating design and balancing in complex multi-agent environments.

cs.LG↗

arXiv2025

Subsecond 3D Mesh Generation for Robot Manipulation

3D meshes are a fundamental representation widely used in computer science and engineering. In robotics, they are particularly valuable because they capture objects in a form that aligns directly with how robots interact with the physical world, enabling core capabilities such as predicting stable grasps, detecting collisions, and simulating dynamics. Although automatic 3D mesh generation methods have shown promising progress in recent years, potentially offering a path toward real-time robot perception, two critical challenges remain. First, generating high-fidelity meshes is prohibitively slow for real-time use, often requiring tens of seconds per object. Second, mesh generation by itself is insufficient. In robotics, a mesh must be contextually grounded, i.e., correctly segmented from the scene and registered with the proper scale and pose. Additionally, unless these contextual grounding steps remain efficient, they simply introduce new bottlenecks. In this work, we introduce an end-to-end system that addresses these challenges, producing a high-quality, contextually grounded 3D mesh from a single RGB-D image in under one second. Our pipeline integrates open-vocabulary object segmentation, accelerated diffusion-based mesh generation, and robust point cloud registration, each optimized for both speed and accuracy. We demonstrate its effectiveness in a real-world manipulation task, showing that it enables meshes to be used as a practical, on-demand representation for robotics perception and planning.

cs.RO↗

arXiv2025

Mesh RAG: Retrieval Augmentation for Autoregressive Mesh Generation

3D meshes are a critical building block for applications ranging from industrial design and gaming to simulation and robotics. Traditionally, meshes are crafted manually by artists, a process that is time-intensive and difficult to scale. To automate and accelerate this asset creation, autoregressive models have emerged as a powerful paradigm for artistic mesh generation. However, current methods to enhance quality typically rely on larger models or longer sequences that result in longer generation time, and their inherent sequential nature imposes a severe quality-speed trade-off. This sequential dependency also significantly complicates incremental editing. To overcome these limitations, we propose Mesh RAG, a novel, training-free, plug-and-play framework for autoregressive mesh generation models. Inspired by RAG for language models, our approach augments the generation process by leveraging point cloud segmentation, spatial transformation, and point cloud registration to retrieve, generate, and integrate mesh components. This retrieval-based approach decouples generation from its strict sequential dependency, facilitating efficient and parallelizable inference. We demonstrate the wide applicability of Mesh RAG across various foundational autoregressive mesh generation models, showing it significantly enhances mesh quality, accelerates generation speed compared to sequential part prediction, and enables incremental editing, all without model retraining.

cs.CV↗

arXiv2025

Hybrid Diffusion Policies with Projective Geometric Algebra for Efficient Robot Manipulation Learning

Diffusion policies are a powerful paradigm for robot learning, but their training is often inefficient. A key reason is that networks must relearn fundamental spatial concepts, such as translations and rotations, from scratch for every new task. To alleviate this redundancy, we propose embedding geometric inductive biases directly into the network architecture using Projective Geometric Algebra (PGA). PGA provides a unified algebraic framework for representing geometric primitives and transformations, allowing neural networks to reason about spatial structure more effectively. In this paper, we introduce hPGA-DP, a novel hybrid diffusion policy that capitalizes on these benefits. Our architecture leverages the Projective Geometric Algebra Transformer (P-GATr) as a state encoder and action decoder, while employing established U-Net or Transformer-based modules for the core denoising process. Through extensive experiments and ablation studies in both simulated and real-world environments, we demonstrate that hPGA-DP significantly improves task performance and training efficiency. Notably, our hybrid approach achieves substantially faster convergence compared to both standard diffusion policies and architectures that rely solely on P-GATr. The project website is available at: https://apollo-lab-yale.github.io/26-ICRA-hPGA-website/.

cs.RO↗

arXiv2025

PRISM-DP: Spatial Pose-based Observations for Diffusion-Policies via Segmentation, Mesh Generation, and Pose Tracking

Diffusion policies generate robot motions by learning to denoise action-space trajectories conditioned on observations. These observations are commonly streams of RGB images, whose high dimensionality includes substantial task-irrelevant information, requiring large models to extract relevant patterns. In contrast, using structured observations like the spatial poses of key objects enables training more compact policies with fewer parameters. However, obtaining accurate object poses in open-set, real-world environments remains challenging, as 6D pose estimation and tracking methods often depend on markers placed on objects beforehand or pre-scanned object meshes that require manual reconstruction. We propose PRISM-DP, an approach that leverages segmentation, mesh generation, and pose tracking models to enable compact diffusion policy learning directly from the spatial poses of task-relevant objects. Crucially, by using a mesh generation model, PRISM-DP eliminates the need for manual mesh creation, improving scalability in open-set environments. Experiments in simulation and the real world show that PRISM-DP outperforms high-dimensional image-based policies and achieves performance comparable to policies trained with ground-truth state information.

cs.RO↗

arXiv2025

Dynamic Rank Adjustment in Diffusion Policies for Efficient and Flexible Training

Diffusion policies trained via offline behavioral cloning have recently gained traction in robotic motion generation. While effective, these policies typically require a large number of trainable parameters. This model size affords powerful representations but also incurs high computational cost during training. Ideally, it would be beneficial to dynamically adjust the trainable portion as needed, balancing representational power with computational efficiency. For example, while overparameterization enables diffusion policies to capture complex robotic behaviors via offline behavioral cloning, the increased computational demand makes online interactive imitation learning impractical due to longer training time. To address this challenge, we present a framework, called DRIFT, that uses the Singular Value Decomposition to enable dynamic rank adjustment during diffusion policy training. We implement and demonstrate the benefits of this framework in DRIFT-DAgger, an imitation learning algorithm that can seamlessly slide between an offline bootstrapping phase and an online interactive phase. We perform extensive experiments to better understand the proposed framework, and demonstrate that DRIFT-DAgger achieves improved sample efficiency and faster training with minimal impact on model performance. The project website is available at: https://apollo-lab-yale.github.io/25-RSS-DRIFT-website/.

cs.RO↗

arXiv2024

Optimizing Active Perception for Learning Simultaneous Viewpoint Selection and Manipulation with Diffusion Policy

Robotic manipulation tasks often rely on static cameras for perception, which can limit flexibility, particularly in scenarios like robotic surgery and cluttered environments where mounting static cameras is impractical. Ideally, robots could jointly learn a policy for dynamic viewpoint and manipulation. However, dynamic viewpoint control requires additional degrees of freedom and intricate coordination with manipulation, which results in more challenging policy learning than single-arm manipulation. To address this complexity, we propose an integrated learning framework that combines diffusion policy with a novel look-at inverse kinematics solver for active perception. Our framework helps better coordinating between perception and manipulation. It automatically optimizes camera orientation for viewpoint selection, while allowing the policy to focus on essential manipulation and positioning decisions. We demonstrate that our integrated approach achieves superior performance and learning efficiency compared to directly applying diffusion policies to configuration space or end-effector space with various rotation representations. Further analysis suggests that these performance differences are driven by inherent variations in the high-frequency components across different state-action spaces.

cs.RO↗

arXiv2024

Developing Path Planning with Behavioral Cloning and Proximal Policy Optimization for Path-Tracking and Static Obstacle Nudging

In autonomous driving, end-to-end methods utilizing Imitation Learning (IL) and Reinforcement Learning (RL) are becoming more and more common. However, they do not involve explicit reasoning like classic robotics workflow and planning with horizons, resulting in strategies implicit and myopic. In this paper, we introduce a path planning method that uses Behavioral Cloning (BC) for path-tracking and Proximal Policy Optimization (PPO) for static obstacle nudging. It outputs lateral offset values to adjust the given reference waypoints and performs modified path for different controllers. Experimental results show that the algorithm can do path following that mimics the expert performance of path-tracking controllers, and avoid collision to fixed obstacles. The method makes a good attempt at planning with learning-based methods in path planning problems of autonomous driving.

cs.RO↗

arXiv2023

Deep Learning for Optimization of Trajectories for Quadrotors

This paper presents a novel learning-based trajectory planning framework for quadrotors that combines model-based optimization techniques with deep learning. Specifically, we formulate the trajectory optimization problem as a quadratic programming (QP) problem with dynamic and collision-free constraints using piecewise trajectory segments through safe flight corridors [1]. We train neural networks to directly learn the time allocation for each segment to generate optimal smooth and fast trajectories. Furthermore, the constrained optimization problem is applied as a separate implicit layer for backpropagation in the network, for which the differential loss function can be obtained. We introduce an additional penalty function to penalize time allocations which result in solutions that violate the constraints to accelerate the training process and increase the success rate of the original optimization problem. To this end, we enable a flexible number of sequences of piece-wise trajectories by adding an extra end-of-sentence token during training. We illustrate the performance of the proposed method via extensive simulation and experimentation and show that it works in real time in diverse, cluttered environments.

cs.RO↗

arXiv2023

MEGA-DAgger: Imitation Learning with Multiple Imperfect Experts

Imitation learning has been widely applied to various autonomous systems thanks to recent development in interactive algorithms that address covariate shift and compounding errors induced by traditional approaches like behavior cloning. However, existing interactive imitation learning methods assume access to one perfect expert. Whereas in reality, it is more likely to have multiple imperfect experts instead. In this paper, we propose MEGA-DAgger, a new DAgger variant that is suitable for interactive learning with multiple imperfect experts. First, unsafe demonstrations are filtered while aggregating the training data, so the imperfect demonstrations have little influence when training the novice policy. Next, experts are evaluated and compared on scenarios-specific metrics to resolve the conflicted labels among experts. Through experiments in autonomous racing scenarios, we demonstrate that policy learned using MEGA-DAgger can outperform both experts and policies learned using the state-of-the-art interactive imitation learning algorithms such as Human-Gated DAgger. The supplementary video can be found at \url{https://youtu.be/wPCht31MHrw}.

cs.LG↗

arXiv2022

A Benchmark Comparison of Imitation Learning-based Control Policies for Autonomous Racing

Autonomous racing with scaled race cars has gained increasing attention as an effective approach for developing perception, planning and control algorithms for safe autonomous driving at the limits of the vehicle's handling. To train agile control policies for autonomous racing, learning-based approaches largely utilize reinforcement learning, albeit with mixed results. In this study, we benchmark a variety of imitation learning policies for racing vehicles that are applied directly or for bootstrapping reinforcement learning both in simulation and on scaled real-world environments. We show that interactive imitation learning techniques outperform traditional imitation learning methods and can greatly improve the performance of reinforcement learning policies by bootstrapping thanks to its better sample efficiency. Our benchmarks provide a foundation for future research on autonomous racing using Imitation Learning and Reinforcement Learning.

cs.RO↗

arXiv2022

Multi-Agent Exploration of an Unknown Sparse Landmark Complex via Deep Reinforcement Learning

In recent years Landmark Complexes have been successfully employed for localization-free and metric-free autonomous exploration using a group of sensing-limited and communication-limited robots in a GPS-denied environment. To ensure rapid and complete exploration, existing works make assumptions on the density and distribution of landmarks in the environment. These assumptions may be overly restrictive, especially in hazardous environments where landmarks may be destroyed or completely missing. In this paper, we first propose a deep reinforcement learning framework for multi-agent cooperative exploration in environments with sparse landmarks while reducing client-server communication. By leveraging recent development on partial observability and credit assignment, our framework can train the exploration policy efficiently for multi-robot systems. The policy receives individual rewards from actions based on a proximity sensor with limited range and resolution, which is combined with group rewards to encourage collaborative exploration and construction of the Landmark Complex through observation of 0-, 1- and 2-dimensional simplices. In addition, we employ a three-stage curriculum learning strategy to mitigate the reward sparsity by gradually adding random obstacles and destroying random landmarks. Experiments in simulation demonstrate that our method outperforms the state-of-the-art landmark complex exploration method in efficiency among different environments with sparse landmarks.

cs.RO↗

arXiv2022

Drive Right: Shaping Public's Trust, Understanding, and Preference Towards Autonomous Vehicles Using a Virtual Reality Driving Simulator

Autonomous vehicles are increasingly introduced into our lives. Yet, people's misunderstanding and mistrust have become the major obstacles to the use of these technologies. In response to this problem, proper work must be done to increase public's understanding and awareness and help drivers rationally evaluate the system. The method proposed in this paper is a virtual reality driving simulator which serves as a low-cost platform for autonomous vehicle demonstration and education. To test the validity of the platform, we recruited 36 participants and conducted a test training drive using three different scenarios. The results show that our simulator successfully increased participants' understanding while favorably changing their attitude towards the autonomous system. The methodology and findings presented in this paper can be further explored by driving schools, auto manufacturers, and policy makers, to improve training for autonomous vehicles.

cs.HC↗