We’re betting on simulation to unlock the next era of robot learning.
Introducing @outworldlabs. We’re building a simulation platform for training and evaluating robot foundation models.
Collecting real-world data and testing robots is expensive and difficult to scale. Simulation offers a path forward, but it needs to reproduce real-world behavior and be easier to use.
We’re building a new physics engine for dexterous manipulation, with tools to create environments, reconstruct and augment real-world data, and scale training and evaluation in one platform.
Working on robot learning? We’d love to chat.
@kimtaehyeon610 We've written our own solvers. We validate them by comparing (in sim vs real):
- task success rates and failure modes
- distributions of physical measurements, like contact forces and object trajectories