Qi-Ze Yu

Autonomous driving VLAs can achieve low-latency, SOTA planning by generating actions before CoT, using understanding solely to supervise generation.

Counterfactual outcomes reveal that existing embodied agents miss 20% of potential success by not exploring alternative skill branches.

A unified architecture achieves SOTA embodied AI by jointly learning physical reasoning, world generation, and action prediction, outperforming prior models in generalization and long-horizon tasks.

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.

Visual grounding can now be achieved at flash speed with precise localization by treating it as parallel evidence extraction, not sequential generation.

X-AuT, a progressive framework that selects layer combinations through short behavioral probes and restores the pruned model through representation alignment, cross-scale distillation, scheduled student-policy supervision, and LoRA finetuning, is introduced.

Video generative priors only translate into robust physical control when paired with explicit world-to-action information routing and synchronized joint denoising rather than standard monolithic fine-tuning.

Motion-aligned latent dynamics enable robots to learn actionable behaviors from human videos without losing visual fidelity across different embodiments.

Trustworthiness in embodied intelligence isn't just about performance; it's about managing risk across a multi-layered framework that ensures safety and reliability in real-world applications.

AffordanceVLA transforms robotic manipulation by using structured affordance cues to create precise perception-action mappings, outperforming traditional models.