Cutting-edge AI research to help our customers stay ahead of the curve
Covariant was founded on research that fundamentally changed the world of AI. Today, we continue to actively research new and novel approaches that we apply to solve the challenges that our robotic systems see in warehouses and fulfillment centers around the world. What normally can take 5-8 years to make it from the lab to the real world, we do in a matter of weeks and months. Learn more about our research here.
While most of our R&D efforts are kept private, giving us a leading edge in providing autonomous capabilities, we do share examples of our advances with the public to help illustrate why these investments matter.
This post explores one of the ways we make this cutting-edge research possible: our commitment to best-in-class engineering practices. We have recently invested in restructuring our AI infrastructure stack to operate in the cloud, using innovative technologies such as SkyPilot, a cloud-based AI infrastructure platform. Take a look at what this means for our growing global customer base.
Best-in-class engineering practices produce best-in-class performance
By leveraging the latest cloud-based AI infrastructure technology, Covariant delivers AI model updates 4x faster to our robots in customers’ warehouses.
Pioneering AI Robotics: In 2017, our founders left OpenAI and UC Berkeley to build the first commercially available foundation model for AI-powered robotic picking — the Covariant Brain.
- Foundation models have transformed AI in the digital world — think ChatGPT for text and Midjourney for images.
- Our world-renowned AI researchers and engineers are taking the same foundation model approach with the Covariant Brain, helping robots understand their surroundings, make informed decisions, and adapt to changing circumstances.
- Covariant Brain-powered robots automate a variety of manual tasks by navigating their environment intelligently and autonomously to increase efficiency, reduce costs, and improve reliability.
- To handle the longtail of challenges seen in real-world scenarios, our AI model is trained data from connected robots in warehouses around the world.
Trained on large amounts of real-world data: AI for language uses vast amounts of online text to train. However, there has not been a pre-existing dataset that represents how robots should interact with the physical world. Real-world deployments are the most ideal learning environment for AI models in the physical world.
- Covariant robots in warehouses in 15 different countries across 4 continents perform millions of picks per month across hundreds of thousands of unique SKUs and items.
- The real-world data from these connected robots is used to continuously train and improve our AI models.
- As the number of our deployments has grown, the amount of real-world data the Covariant Brain processes has also grown massively.
Leveraging the cloud: To handle this increase in data, we scaled up our AI infrastructure for computation and processing beyond what our on-premise infrastructure could provide.
Scaling infrastructure to process more data, faster: With cloud-based AI infrastructure from SkyPilot, we now deliver AI model updates 4 times faster to Covariant robots in your warehouse.
- Using more compute power on the latest generation of GPUs:
- Increased processors from dozens of GPUs to 1000’s of GPUs.
- Able to run 100’s of computing jobs at the same time, instead of just a handful.
- Our global AI research and engineering teams can run more experiments simultaneously.
- Your Covariant robots learn from other robots faster, helping further increase your operational efficiency.
- Covariant Brain AI models train and update 4x faster.
- Larger models, using orders of magnitude more data, keep increasing picking accuracy and speed.
Delivering continuous innovation: As the market keeps evolving, we help our customers stay ahead of the competition by inventing new methodologies and leveraging the latest cutting-edge technologies.