There's a growing disconnect between publication-oriented research into learning systems and the needs of real-world customers with scenarios that demand ever-more autonomous and flexible automation solutions. At Covariant, we try to be the world’s “hands” — providing automation solutions from de-palletizing to apparel induction. In the real world, you have to deal with a very long tail of challenges, and the AI solutions necessary for dealing with them at the highest levels of reliability need to go beyond the latest and greatest in academic papers.

The pillars behind our approach to research at Covariant are:

  • Composability of learned models that gives us interpretability and flexibility to master new problem domains quickly.
  • Clever solutions to maximize the amount of self/semi-supervised learning, allowing us to best utilize our exponentially growing dataset of object SKUs and robotic-scene interactions.
  • Ever-more realistic simulated environments that provide state-of-the-art sim-to-real transfer.
  • Real-time nature of robotics requires investment in improving latency without sacrificing model capacity.

Hopefully, by the end of this session, you'll appreciate both the technical challenges and the exciting commercial possibilities of AI Robotics.

Sign up for our newsletter

Get the latest news in AI Robotics, picked just for you

Related RFM-1 update: In-context learning to improve grasping