Real-to-sim robotics data and simulation service
Convert sparse captures of real-world robotics environments into simulated environments, randomize conditions within those environments, and use the resulting data to train and evaluate robotic policies before deployment.
From a16z — Why World Models Could Change Robotics, 3D, and Creativity at 24:05
Problem: Collecting and densely reconstructing real-world environments for robotics simulation is slow, laborious, and a major bottleneck because robotic policies must be exposed to many possible conditions and failures before deployment.
For: Robotics teams training robots for industrial tasks such as cable handling or dishwashing and needing more real-world variation and simulation data.
Products from this video
Examples
- Industrial cable handling: a robotic arm needs training data, policy evaluation, and deployment in a cable-handling environment; randomization can vary cable bends and other conditions.