arXiv subjects

Meibao Yao

Publications and source records attributed to Meibao Yao.

4 recordsLinked to original sources

arXiv2026

RetroHolmes: When Semantic Plausibility Fails Retrospective Physical Process Reasoning

Humans can infer hidden physical processes from sparse observations, yet current evaluation protocols for Vision Language Models fail to assess whether such physical reasoning is genuinely captured. To address this gap, we introduce Retrospective Physical Process Reasoning, a new evaluation paradigm to reason backward from outcomes under explicit physical constraints. Building on the paradigm, we present RetroHolmes, the first real-world benchmark for Retrospective Physical Process Reasoning, comprising object-centric image pairs annotated with reachability labels and causal step sequences across diverse physical transitions. Using RetroHolmes, we analyze state of the art Vision Language Models and uncover systematic failure modes, including judgment bias in reachability assessment and belief dominance over physical evidence, mirroring sycophancy behavior observed in large language models. We further demonstrate a simple analysis-by-synthesis instantiation with visual simulation as an intermediate step, validating the diagnostic value of RetroHolmes and highlighting the importance of physically grounded intermediate representations for physical reasoning.

cs.MM↗

arXiv2025

Beyond Success: Refining Elegant Robot Manipulation from Mixed-Quality Data via Just-in-Time Intervention

Vision-Language-Action (VLA) models have enabled notable progress in general-purpose robotic manipulation, yet their learned policies often exhibit variable execution quality. We attribute this variability to the mixed-quality nature of human demonstrations, where the implicit principles that govern how actions should be carried out are only partially satisfied. To address this challenge, we introduce the LIBERO-Elegant benchmark with explicit criteria for evaluating execution quality. Using these criteria, we develop a decoupled refinement framework that improves execution quality without modifying or retraining the base VLA policy. We formalize Elegant Execution as the satisfaction of Implicit Task Constraints (ITCs) and train an Elegance Critic via offline Calibrated Q-Learning to estimate the expected quality of candidate actions. At inference time, a Just-in-Time Intervention (JITI) mechanism monitors critic confidence and intervenes only at decision-critical moments, providing selective, on-demand refinement. Experiments on LIBERO-Elegant and real-world manipulation tasks show that the learned Elegance Critic substantially improves execution quality, even on unseen tasks. The proposed model enables robotic control that values not only whether tasks succeed, but also how they are performed.

cs.RO↗

arXiv2025

GCNT: Graph-Based Transformer Policies for Morphology-Agnostic Reinforcement Learning

Training a universal controller for robots with different morphologies is a promising research trend, since it can significantly enhance the robustness and resilience of the robotic system. However, diverse morphologies can yield different dimensions of state space and action space, making it difficult to comply with traditional policy networks. Existing methods address this issue by modularizing the robot configuration, while do not adequately extract and utilize the overall morphological information, which has been proven crucial for training a universal controller. To this end, we propose GCNT, a morphology-agnostic policy network based on improved Graph Convolutional Network (GCN) and Transformer. It exploits the fact that GCN and Transformer can handle arbitrary number of modules to achieve compatibility with diverse morphologies. Our key insight is that the GCN is able to efficiently extract morphology information of robots, while Transformer ensures that it is fully utilized by allowing each node of the robot to communicate this information directly. Experimental results show that our method can generate resilient locomotion behaviors for robots with different configurations, including zero-shot generalization to robot morphologies not seen during training. In particular, GCNT achieved the best performance on 8 tasks in the 2 standard benchmarks.

cs.RO↗

arXiv2024

Passive None-line-of-sight imaging with arbitrary scene condition and detection pattern in small amount of prior data

Passive Non-Line-of-Sight (NLOS) imaging requires to reconstruct objects which cannot be seen in line without using external controllable light sources. It can be widely applied in areas like counter-terrorism, urban-Warfare, autonomous-driving and robot-vision. Existing methods for passive NLOS typically required extensive prior information and significant computational resources to establish light transport matrices or train neural networks. These constraints pose significant challenges for transitioning models to different NLOS scenarios. Thus, the pressing issue in passive NLOS imaging currently lies in whether it is possible to estimate the light transport matrices which corresponding to relay surfaces and scenes, as well as the specific distribution of targets, with a small amount of prior knowledge. In this work, we hypothesized a high-dimensional manifold and mathematically proved its existence. Within this high-dimensional manifold, the structural information of obscured targets is minimally disrupted. Therefore, we proposed a universal framework named High-Dimensional Projection Selection (HDPS) which can establish this high-dimensional manifold and output its projection onto corresponding surfaces on low-dimensional. HDPS can be applied to most mature network architectures and estimate the distribution of target and light spot obtained by camera with only minimal prior data. Certainly, with the help of the estimated information, it can establish a high-dimensional manifold consisting of target and input. As demonstrated in experiment, our framework, even when applied to the most basic network structures, can achieve higher accuracy results with significantly smaller amounts of prior data. Thereby, our approach enables passive NLOS scenarios to reconstruct target by limited prior data and computational resources.

physics.optics↗