Wei-Yun Yau

Predicting compact, learned future representations for world-action modeling slashes training costs by 65x and outperforms native visual targets, even with smaller models.

Anchoring a UAV to its environment can reduce positional drift by over 95%, enabling reliable performance in contact-based tasks even in windy conditions.

A 99% reduction in token count without sacrificing spatial and semantic integrity leads to a 22.8% boost in real-world robotic task success rates.

EAGOR revolutionizes embodied reasoning by directly leveraging the spherical nature of 360-degree observations, leading to remarkable gains in navigation accuracy and efficiency.

Roken can generate coordinated multi-robot trajectories in real-time, achieving higher success rates than traditional sequential planning methods.