No Priors Ep. 48 | With Covariant CEO Peter Chen
Covariant co-founder and CEO Peter Chen joins Sarah on No Priors to discuss building robotics foundation models, deploying robots in warehouses, and the data and safety challenges involved.
In this talk
- 1:02
Why Covariant started with customer deployments
Chen was drawn to robotics because it combines learning from large datasets with systems that make decisions and act in the physical world. Covariant started in 2017 with the belief that robotics foundation models would need broad data, which required fleets of robots delivering value to customers. He describes an incremental path: ship useful systems, learn from production, and use that learning to expand what the models can do.
- 8:14
Making warehouse robots more adaptable
Chen says most deployed robots repeat preprogrammed motions and cannot adapt to changing circumstances. Covariant focuses on warehouse manipulation, where robots must identify, grasp, and route diverse items. He uses a put wall as an example: a robot sorts orders by picking items such as phones and sending each to the right customer. The company reported about 200 employees and deployments across more than 10 countries on three continents.
- 19:50
Grounding models in physical interaction
Chen explains that internet image and text pairs can connect concepts to visible objects, but often lack the precision needed for robotic manipulation. Robots also need data linking actions to outcomes, such as how force affects a tomato or whether a grasp succeeds. Covariant's central bet is that robotics progress depends on collecting real-world data; Chen sees simulation as useful for augmentation, but difficult for modeling contact and the many distinct items in warehouses.
- 32:53
Reliability, deployment, and safety
Chen says a robotics breakthrough requires both broad capability and high reliability, since a 30 percent rate of catastrophic outcomes would be unacceptable. He expects industrial uses to arrive before home robots because equipment can be used continuously and justify its cost. He imagines people overseeing fleets of robots, and says Covariant follows established industrial safety requirements such as safety cages or certified controllers.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.