Preload

Teams training robots or foundation models on physical tasks rely on video, which only captures a flat projection of the world. Video cannot record the force, contact, and motion involved in how a human hand holds, presses, or releases an object. This means better model architectures cannot recover information that was never measured in the first place, limiting how well models can learn real physical skills from demonstrations.

Preload builds hardware and software to capture human skill in a measurable, synchronized form and turn it into robot capability. Its hardware stack includes a head-mounted camera for point-of-view video, a tactile sensing glove that measures pressure across fingers and palm, and an EMG band that reads forearm muscle activation during a grip. This synchronized video, touch, and force data feeds a five-stage pipeline: sense, collect, train, evaluate, and deploy, moving from capturing human demonstrations to training robot policies, testing them against ground truth task targets, and deploying validated skills on real-world tasks.