Kei Okada

A biomimetic skin can achieve joint angle estimation with remarkable accuracy, outperforming traditional muscle sensing alone.

Achieving robust brachiation on a life-sized robot reveals how sparse waypoint guidance can enhance complex motion learning in challenging environments.

Sharing actuators across reconfigurable robotic modules could revolutionize the design and deployment of multi-functional robots.

Gated VLA-Cache recovers lost accuracy in real-time control while slashing compute costs by leveraging model uncertainty.

VLAff transforms human video insights into actionable robot manipulation skills, achieving state-of-the-art performance while enabling zero-shot learning capabilities.

Wrench-based exploration can significantly enhance learning efficiency in legged robots, but misalignment with physical structure may lead to ineffective behaviors.

MEVION can perform high-speed and high-force manipulations, enabling data collection for complex tasks that were previously impossible with existing systems.

Joint receptors, often overshadowed by muscle spindles, can enable surprisingly accurate proprioception on their own.

By automatically generating fingertip workspace clouds, FSG enables real-time, human-like grasp generation for robotic hands with arbitrary structures, sidestepping the inverse kinematics bottleneck.