Synapath AI builds data infrastructure that converts human work videos and real-world interaction into robot-training trajectories. Its pipeline combines video reconstruction, behavior representation and cross-embodiment mapping for foundation-model and robot-learning teams.
About
Synapath AI develops data infrastructure for embodied-AI and robot-learning teams. Its pipeline reconstructs human motion and physical interaction from video, converts those demonstrations into structured trajectories, and retargets the resulting data to different robot embodiments. This approach is intended to expand training data beyond demonstrations collected through direct robot teleoperation, while preserving contact and action information needed for policy learning.
The HORA dataset contains more than 150,000 trajectories extracted from real-world human video and adds multimodal representations for embodied training. RoboWheel extends the pipeline to cross-embodiment, high-fidelity physical-interaction data that can be replayed on robot arms, dexterous hands and humanoid platforms. Synapath supplies reconstructed datasets and the data engine used to create them. Robot manufacturers and research teams remain responsible for model training, hardware integration, safety validation and deployment of the resulting policies.