Xi Chen

Explicitly grounding the goal object in view unlocks near-expert performance in object-goal navigation, revealing perception rather than policy as the key limitation.

A simple test-time trigger and latent world model can boost VLA policy success by 15% without retraining, by intelligently selecting the best action from stochastic outputs.

Simply handing VLMs a 360° field of view fails to improve embodied navigation unless paired with dynamic multi-step action planning and explicit geometric representation, which together yield an 11.9% jump over SOTA.

A Retrieval-Augmented Generation framework for scientific image quality assessment, designed to simultaneously address both the understanding track (SIQA-U) and the scoring track (SIQA-S) of the SIQA challenge, is proposed.

This paper proposes and analyzes Comparison-based Preference Optimization (ComPO), a zeroth-order alignment method based on comparison oracles, and establishes a convergence guarantee for its basic offline scheme under smoothness, gradient sparsity, and compatibility between the oracle and a latent objective.