arXiv subjects

Jiajun Wu

Publications and source records attributed to Jiajun Wu.

At least 127 records · Page 7Linked to original sources

arXiv2025

ENACT: Evaluating Embodied Cognition with World Modeling of Egocentric Interaction

Embodied cognition argues that intelligence arises from sensorimotor interaction rather than passive observation. It raises an intriguing question: do modern vision-language models (VLMs), trained largely in a disembodied manner, exhibit signs of embodied cognition? We introduce ENACT, a benchmark that casts evaluation of embodied cognition as world modeling from egocentric interaction in a visual question answering (VQA) format. Framed as a partially observable Markov decision process (POMDP) whose actions are scene graph changes, ENACT comprises two complementary sequence reordering tasks: forward world modeling (reorder shuffled observations given actions) and inverse world modeling (reorder shuffled actions given observations). While conceptually simple, solving these tasks implicitly demands capabilities central to embodied cognition-affordance recognition, action-effect reasoning, embodied awareness, and interactive, long-horizon memory from partially observable egocentric input, while avoiding low-level image synthesis that could confound the evaluation. We provide a scalable pipeline that synthesizes QA pairs from robotics simulation (BEHAVIOR) and evaluates models on 8,972 QA pairs spanning long-horizon home-scale activities. Experiments reveal a performance gap between frontier VLMs and humans that widens with interaction horizon. Models consistently perform better on the inverse task than the forward one and exhibit anthropocentric biases, including a preference for right-handed actions and degradation when camera intrinsics or viewpoints deviate from human vision. Website at https://enact-embodied-cognition.github.io/.

cs.AI↗

arXiv2025

Model-Based Policy Adaptation for Closed-Loop End-to-End Autonomous Driving

End-to-end (E2E) autonomous driving models have demonstrated strong performance in open-loop evaluations but often suffer from cascading errors and poor generalization in closed-loop settings. To address this gap, we propose Model-based Policy Adaptation (MPA), a general framework that enhances the robustness and safety of pretrained E2E driving agents during deployment. MPA first generates diverse counterfactual trajectories using a geometry-consistent simulation engine, exposing the agent to scenarios beyond the original dataset. Based on this generated data, MPA trains a diffusion-based policy adapter to refine the base policy's predictions and a multi-step Q value model to evaluate long-term outcomes. At inference time, the adapter proposes multiple trajectory candidates, and the Q value model selects the one with the highest expected utility. Experiments on the nuScenes benchmark using a photorealistic closed-loop simulator demonstrate that MPA significantly improves performance across in-domain, out-of-domain, and safety-critical scenarios. We further investigate how the scale of counterfactual data and inference-time guidance strategies affect overall effectiveness.

cs.RO↗

arXiv2025

LLMC+: Benchmarking Vision-Language Model Compression with a Plug-and-play Toolkit

Large Vision-Language Models (VLMs) exhibit impressive multi-modal capabilities but suffer from prohibitive computational and memory demands, due to their long visual token sequences and massive parameter sizes. To address these issues, recent works have proposed training-free compression methods. However, existing efforts often suffer from three major limitations: (1) Current approaches do not decompose techniques into comparable modules, hindering fair evaluation across spatial and temporal redundancy. (2) Evaluation confined to simple single-turn tasks, failing to reflect performance in realistic scenarios. (3) Isolated use of individual compression techniques, without exploring their joint potential. To overcome these gaps, we introduce LLMC+, a comprehensive VLM compression benchmark with a versatile, plug-and-play toolkit. LLMC+ supports over 20 algorithms across five representative VLM families and enables systematic study of token-level and model-level compression. Our benchmark reveals that: (1) Spatial and temporal redundancies demand distinct technical strategies. (2) Token reduction methods degrade significantly in multi-turn dialogue and detail-sensitive tasks. (3) Combining token and model compression achieves extreme compression with minimal performance loss. We believe LLMC+ will facilitate fair evaluation and inspire future research in efficient VLM. Our code is available at https://github.com/ModelTC/LightCompress.

cs.CV↗

arXiv2025

Riemann-Silberstein geometric phase in 4D polarization space

Geometric phase is a far-reaching concept in quantum and classical physics. The first discovered geometric phase, the Pancharatnam-Berry (PB) phase, has profoundly shaped nanophotonics through metasurfaces. However, the PB phase arises from SU(2) polarization evolution and is constrained to a 2D polarization space, failing to capture the full polarization degrees of freedom. We generalize geometric phase to the 4D Riemann-Silberstein (RS) space that simultaneously describes electric, magnetic, and hybrid electric-magnetic polarizations. We show that SU(4) polarization evolution can generate a new geometric phase, the RS phase, alongside the PB phase. Unlike the PB phase that typically manifests in circularly polarized light, the RS phase can emerge in arbitrarily polarized light. Together, they enable a high-dimensional geometric framework for light propagation across general interfaces. We reveal that the phase shifts governed by Fresnel equations are direct manifestations of the RS-space geometric phases, integrating a century-old wave theory into this paradigm. We experimentally validate the framework using metasurfaces and achieve high-dimensional wavefront manipulation. Our work offers fundamental insights into the geometric nature of light-matter interactions, with implications for topological and non-Abelian physics in classical wave systems.

physics.optics↗

arXiv2025

VisualMimic: Visual Humanoid Loco-Manipulation via Motion Tracking and Generation

Humanoid loco-manipulation in unstructured environments demands tight integration of egocentric perception and whole-body control. However, existing approaches either depend on external motion capture systems or fail to generalize across diverse tasks. We introduce VisualMimic, a visual sim-to-real framework that unifies egocentric vision with hierarchical whole-body control for humanoid robots. VisualMimic combines a task-agnostic low-level keypoint tracker -- trained from human motion data via a teacher-student scheme -- with a task-specific high-level policy that generates keypoint commands from visual and proprioceptive input. To ensure stable training, we inject noise into the low-level policy and clip high-level actions using human motion statistics. VisualMimic enables zero-shot transfer of visuomotor policies trained in simulation to real humanoid robots, accomplishing a wide range of loco-manipulation tasks such as box lifting, pushing, football dribbling, and kicking. Beyond controlled laboratory settings, our policies also generalize robustly to outdoor environments. Videos are available at: https://visualmimic.github.io .

cs.RO↗

arXiv2025

10 Open Challenges Steering the Future of Vision-Language-Action Models

Due to their ability of follow natural language instructions, vision-language-action (VLA) models are increasingly prevalent in the embodied AI arena, following the widespread success of their precursors -- LLMs and VLMs. In this paper, we discuss 10 principal milestones in the ongoing development of VLA models -- multimodality, reasoning, data, evaluation, cross-robot action generalization, efficiency, whole-body coordination, safety, agents, and coordination with humans. Furthermore, we discuss the emerging trends of using spatial understanding, modeling world dynamics, post training, and data synthesis -- all aiming to reach these milestones. Through these discussions, we hope to bring attention to the research avenues that may accelerate the development of VLA models into wider acceptability.

cs.RO↗

arXiv2025

TWIST2: Scalable, Portable, and Holistic Humanoid Data Collection System

Large-scale data has driven breakthroughs in robotics, from language models to vision-language-action models in bimanual manipulation. However, humanoid robotics lacks equally effective data collection frameworks. Existing humanoid teleoperation systems either use decoupled control or depend on expensive motion capture setups. We introduce TWIST2, a portable, mocap-free humanoid teleoperation and data collection system that preserves full whole-body control while advancing scalability. Our system leverages PICO4U VR for obtaining real-time whole-body human motions, with a custom 2-DoF robot neck (cost around $250) for egocentric vision, enabling holistic human-to-humanoid control. We demonstrate long-horizon dexterous and mobile humanoid skills and we can collect 100 demonstrations in 15 minutes with an almost 100% success rate. Building on this pipeline, we propose a hierarchical visuomotor policy framework that autonomously controls the full humanoid body based on egocentric vision. Our visuomotor policy successfully demonstrates whole-body dexterous manipulation and dynamic kicking tasks. The entire system is fully reproducible and open-sourced at https://yanjieze.com/TWIST2 . Our collected dataset is also open-sourced at https://twist-data.github.io .

cs.RO↗

arXiv2025

Web-Scale Collection of Video Data for 4D Animal Reconstruction

Computer vision for animals holds great promise for wildlife research but often depends on large-scale data, while existing collection methods rely on controlled capture setups. Recent data-driven approaches show the potential of single-view, non-invasive analysis, yet current animal video datasets are limited--offering as few as 2.4K 15-frame clips and lacking key processing for animal-centric 3D/4D tasks. We introduce an automated pipeline that mines YouTube videos and processes them into object-centric clips, along with auxiliary annotations valuable for downstream tasks like pose estimation, tracking, and 3D/4D reconstruction. Using this pipeline, we amass 30K videos (2M frames)--an order of magnitude more than prior works. To demonstrate its utility, we focus on the 4D quadruped animal reconstruction task. To support this task, we present Animal-in-Motion (AiM), a benchmark of 230 manually filtered sequences with 11K frames showcasing clean, diverse animal motions. We evaluate state-of-the-art model-based and model-free methods on Animal-in-Motion, finding that 2D metrics favor the former despite unrealistic 3D shapes, while the latter yields more natural reconstructions but scores lower--revealing a gap in current evaluation. To address this, we enhance a recent model-free approach with sequence-level optimization, establishing the first 4D animal reconstruction baseline. Together, our pipeline, benchmark, and baseline aim to advance large-scale, markerless 4D animal reconstruction and related tasks from in-the-wild videos. Code and datasets are available at https://github.com/briannlongzhao/Animal-in-Motion.

cs.CV↗

arXiv2025

Exploringand Unleashing the Power of Large Language Models in CI/CD Configuration Translation

Continuous Integration (CI) is a cornerstone of modern collaborative software development, and numerous CI platforms are available. Differences in maintenance overhead, reliability, and integration depth with code-hosting platforms make migration between CI platforms a common practice. A central step in migration is translating CI configurations, which is challenging due to the intrinsic complexity of CI configurations and the need to understand semantic differences and relationships across CI platforms. With the advent of large language models (LLMs), recent advances in software engineering highlight their potential for CI configuration translation. In this paper, we present a study on LLM-based CI configuration translation, focusing on the migration from Travis CI to GitHub Actions. First, using 811 migration records, we quantify the effort involved and find that developers read an average of 38 lines of Travis configuration and write 58 lines of GitHub Actions configuration, with nearly half of the migrations requiring multiple commits. We further analyze translations produced by each of the four LLMs and identify 1,121 issues grouped into four categories: logic inconsistencies (38%), platform discrepancies (32%), environment errors (25%), and syntax errors (5%). Finally, we evaluate three enhancement strategies and show that combining guideline-based prompting with iterative refinement achieves the best performance, reaching a Build Success Rate of 75.5%-nearly a threefold improvement over GPT-4o with a basic prompt.

cs.SE↗

arXiv2025

Rotate Both Ways: Time-and-Order RoPE for Generative Recommendation

Generative recommenders, typically transformer-based autoregressive models, predict the next item or action from a user's interaction history. Their effectiveness depends on how the model represents where an interaction event occurs in the sequence (discrete index) and when it occurred in wall-clock time. Prevailing approaches inject time via learned embeddings or relative attention biases. In this paper, we argue that RoPE-based approaches, if designed properly, can be a stronger alternative for jointly modeling temporal and sequential information in user behavior sequences. While vanilla RoPE in LLMs considers only token order, generative recommendation requires incorporating both event time and token index. To address this, we propose Time-and-Order RoPE (TO-RoPE), a family of rotary position embedding designs that treat index and time as angle sources shaping the query-key geometry directly. We present three instantiations: early fusion, split-by-dim, and split-by-head. Extensive experiments on both publicly available datasets and a proprietary industrial dataset show that TO-RoPE variants consistently improve accuracy over existing methods for encoding time and index. These results position rotary embeddings as a simple, principled, and deployment-friendly foundation for generative recommendation.

cs.IR↗

arXiv2025

The Layout Is the Model: On Action-Item Coupling in Generative Recommendation

Generative Recommendation (GR) models treat a user's interaction history as a sequence to be autoregressively predicted. When both items and actions (e.g., watch time, purchase, comment) are modeled, the layout-the ordering and visibility of item/action tokens-critically determines what information the model can use and how it generalizes. We present a unified study of token layouts for GR grounded in first principles: (P1) maximize item/action signal in both input/output space, (P2) preserve the conditioning relationship "action given item" and (P3) no information leakage. While interleaved layout (where item and action occupy separate tokens) naturally satisfies these principles, it also bloats sequence length with larger training/inference cost. On the non-interleaved front, we design a novel and effective approach, Lagged Action Conditioning (LAC), which appears strange on the surface but aligns well with the design principles to yield strong accuracy. Comprehensive experiments on public datasets and large-scale production logs evaluate different layout options and empirically verifies the design principles. Our proposed non-interleaved method, LAC, achieves competitive or superior quality at substantially lower FLOPs than interleaving. Our findings offer actionable guidance for assembling GR systems that are both accurate and efficient.

cs.IR↗

arXiv2025

VAGEN: Reinforcing World Model Reasoning for Multi-Turn VLM Agents

A key challenge in training Vision-Language Model (VLM) agents, compared to Language Model (LLM) agents, lies in the shift from textual states to complex visual observations. This transition introduces partial observability and demands robust world modeling. We ask: Can VLM agents construct internal world models through explicit visual state reasoning? To address this question, we architecturally enforce and reward the agent's reasoning process via reinforcement learning (RL), formulating it as a Partially Observable Markov Decision Process (POMDP). We find that decomposing the agent's reasoning into State Estimation ("what is the current state?") and Transition Modeling ("what comes next?") is critical for success, as demonstrated through five reasoning strategies. Our investigation into how agents represent internal beliefs reveals that the optimal representation is task-dependent: Natural Language excels at capturing semantic relationships in general tasks, while Structured formats are indispensable for precise manipulation and control. Building on these insights, we design a World Modeling Reward that provides dense, turn-level supervision for accurate state prediction, and introduce Bi-Level General Advantage Estimation (Bi-Level GAE) for turn-aware credit assignment. Through this form of visual state reasoning, a 3B-parameter model achieves a score of 0.82 across five diverse agent benchmarks, representing a 3$\times$ improvement over its untrained counterpart (0.21) and outperforming proprietary reasoning models such as GPT-5 (0.75), Gemini 2.5 Pro (0.67) and Claude 4.5 (0.62). All experiments are conducted within our VAGEN framework, a scalable system for training and analyzing multi-turn VLM agents in diverse visual environments. Code and data are publicly available at https://vagen-ai.github.io.

cs.AI↗

arXiv2025

Ctrl-VI: Controllable Video Synthesis via Variational Inference

Many video workflows benefit from a mixture of user controls with varying granularity, from exact 4D object trajectories and camera paths to coarse text prompts, while existing video generative models are typically trained for fixed input formats. We develop Ctrl-VI, a video synthesis method that addresses this need and generates samples with high controllability for specified elements while maintaining diversity for under-specified ones. We cast the task as variational inference to approximate a composed distribution, leveraging multiple video generation backbones to account for all task constraints collectively. To address the optimization challenge, we break down the problem into step-wise KL divergence minimization over an annealed sequence of distributions, and further propose a context-conditioned factorization technique that reduces modes in the solution space to circumvent local optima. Experiments suggest that our method produces samples with improved controllability, diversity, and 3D consistency compared to prior works.

cs.CV↗

arXiv2025

Coupled Diffusion Sampling for Training-Free Multi-View Image Editing

We present an inference-time diffusion sampling method to perform multi-view consistent image editing using pre-trained 2D image editing models. These models can independently produce high-quality edits for each image in a set of multi-view images of a 3D scene or object, but they do not maintain consistency across views. Existing approaches typically address this by optimizing over explicit 3D representations, but they suffer from a lengthy optimization process and instability under sparse view settings. We propose an implicit 3D regularization approach by constraining the generated 2D image sequences to adhere to a pre-trained multi-view image distribution. This is achieved through coupled diffusion sampling, a simple diffusion sampling technique that concurrently samples two trajectories from both a multi-view image distribution and a 2D edited image distribution, using a coupling term to enforce the multi-view consistency among the generated images. We validate the effectiveness and generality of this framework on three distinct multi-view image editing tasks, demonstrating its applicability across various model architectures and highlighting its potential as a general solution for multi-view consistent editing.

cs.CV↗

arXiv2025

Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas

Large Vision Language Models (VLMs) have long struggled with spatial reasoning tasks. Surprisingly, even simple spatial reasoning tasks, such as recognizing "under" or "behind" relationships between only two objects, pose significant challenges for current VLMs. In this work, we study the spatial reasoning challenge from the lens of mechanistic interpretability, diving into the model's internal states to examine the interactions between image and text tokens. By tracing attention distribution over the image through out intermediate layers, we observe that successful spatial reasoning correlates strongly with the model's ability to align its attention distribution with actual object locations, particularly differing between familiar and unfamiliar spatial relationships. Motivated by these findings, we propose ADAPTVIS based on inference-time confidence scores to sharpen the attention on highly relevant regions when confident, while smoothing and broadening the attention window to consider a wider context when confidence is lower. This training-free decoding method shows significant improvement (e.g., up to a 50 absolute point improvement) on spatial reasoning benchmarks such as WhatsUp and VSR with negligible cost. We make code and data publicly available for research purposes at https://github.com/shiqichen17/AdaptVis.

cs.CL↗

arXiv2025

Packaging of Low-Environmental-Sensitivity WGM Resonators for Practical Applications

We present a prism-coupled packaging strategy for whispering-gallery mode resonators (WGMRs). Utilizing an all-solid-state optical adhesive process with active temperature control and hermetic sealing, the package exhibits exceptional long-term stability and environmental robustness. A standalone WGMR module was characterized, demonstrating a temperature sensitivity below 10^-7 /°C and a low-frequency Z-axis acceleration sensitivity below 10^-10 /g. The module was applied as a stable optical frequency reference, achieving a short-term frequency stability of 2x10^-13 at 2 ms, and as a nonlinear photonic platform, generating Kerr soliton microcombs with a pump power of 100 mW. This compact, robust packaging solution enhances the applicability of WGMRs in real-world applications such as narrow-linewidth lasers and portable microcombs, facilitating the transition from laboratory research to practical deployment.

physics.optics↗

arXiv2025

From Programs to Poses: Factored Real-World Scene Generation via Learned Program Libraries

Real-world scenes, such as those in ScanNet, are difficult to capture, with highly limited data available. Generating realistic scenes with varied object poses remains an open and challenging task. In this work, we propose FactoredScenes, a framework that synthesizes realistic 3D scenes by leveraging the underlying structure of rooms while learning the variation of object poses from lived-in scenes. We introduce a factored representation that decomposes scenes into hierarchically organized concepts of room programs and object poses. To encode structure, FactoredScenes learns a library of functions capturing reusable layout patterns from which scenes are drawn, then uses large language models to generate high-level programs, regularized by the learned library. To represent scene variations, FactoredScenes learns a program-conditioned model to hierarchically predict object poses, and retrieves and places 3D objects in a scene. We show that FactoredScenes generates realistic, real-world rooms that are difficult to distinguish from real ScanNet scenes.

cs.CV↗

arXiv2025

Product of Experts for Visual Generation

Modern neural models capture rich priors and have complementary knowledge over shared data domains, e.g., images and videos. Integrating diverse knowledge from multiple sources -- including visual generative models, visual language models, and sources with human-crafted knowledge such as graphics engines and physics simulators -- remains under-explored. We propose a Product of Experts (PoE) framework that performs inference-time knowledge composition from heterogeneous models. This training-free approach samples from the product distribution across experts via Annealed Importance Sampling (AIS). Our framework shows practical benefits in image and video synthesis tasks, yielding better controllability than monolithic methods and additionally providing flexible user interfaces for specifying visual generation goals.

cs.CV↗