Ming-Lei Li

A 4B grounding foundation model achieves state-of-the-art visual grounding, outperforming larger models and boosting downstream robotic and autonomous driving performance by up to 24.8% in OOD settings.

Existing SFT reweighting methods fail to reverse harmful learned features; a new entropy-guided approach enables precise suppression, reversal, and extrapolation of pretrained knowledge.

UMI-Bridge, which uses UMI as an intermediate domain to align representations according to action equivalence rather than pixel similarity, supports action-anchored latent alignment for data-efficient robot learning and UMI-to-robot task transfer.

Training-time Gaussian distillation can elevate WAM performance by over 19% by effectively integrating geometric and semantic information without altering deployment architecture.

Policies may succeed in tasks but still violate deformation tolerances, revealing a critical gap in current evaluation methods for deformable-object manipulation.

Action-conditioned objectives can significantly enhance the effectiveness of Euclidean-cost MPC by aligning latent representations with real task progress.

Zero-robot post-training on high-fidelity UMI data achieves real-world performance that rivals traditional teleoperation methods, revolutionizing data efficiency in robotic manipulation.

Human demonstrations can yield over 10x the recovery data for robots, dramatically enhancing their ability to recover from failures in real-world tasks.

Despite high success rates for isolated skills, robots frequently stall in long-horizon tasks due to semantic handoff failures, revealing a hidden challenge in skill composition.