HumanoidPose Framework Recovers Robot Joint Configurations from Video
October 17, 2026
HumanoidPose combines pretrained human-pose features with robot kinematics to estimate full-body joint configurations from video alone. This allows for state estimation in scenarios where onboard sensor data is restricted due to privacy or communication constraints.
HOW THIS AFFECTS YOU
●
researcherExplore this for training robot policies using unlabelled or privacy-sensitive video data.