Loading patent…
OwnerMichael HEALY, Craig CHENEY +2inventors
InventorsMichael HEALY, Craig CHENEY, Fabrizio SANTINI, Maxim ANTINORI
Estimating the term…
Systems and methods are disclosed for generating training data for robotic artificial intelligence models using augmented reality (AR) glasses. A human operator performs a task while AR sensors capture multimodal data including RGB video, hand and arm pose trajectories, and metadata. Operator trajectories are converted into robot joint trajectories using inverse kinematics applied to parameterized robot models. A data-augmentation pipeline removes the operator using transformer-based segmentation and spatio-temporal in-painting, and inserts a to-scale robot embodiment consistent with the calculated trajectories. Real-time AR feedback enforces data-quality constraints and reduces recollection time. The system produces synchronized datasets, including raw video, augmented robot-inlaid video, and trajectory files that are suitable for training robotic AI models with reduced real-to-sim discrepancy across diverse manipulation tasks.
A method of generating robotic training data for the performance of a task, the method comprising:
Augmented reality glasses for training a robotic system performing a task, the augmented reality glasses comprising:
A method of determining a pose of a robot in real time with augmented reality glasses, the method comprising:
Dependencies are inferred from the claim text.