Memories.ai builds standardized video pipeline to train robots
Memories.ai creates a new data pipeline that turns first-person video into training material for robotics companies.
Memories.ai is working on a new data pipelineData pipelineA set of tools and processes used to collect, process, and move data from one place to another. Open in glossary designed to help robotics companies train their machines. Based in San Francisco, the startup combines standardized video collection with artificial intelligence annotation. This process turns everyday first-person footage into useful trainingTrainingTeaching an AI model by showing it very large amounts of data. Open in glossary data so robots can better understand the physical world. The firm aims to act as a long-term data infrastructure partner rather than a simple data seller.
Standardized Video and Annotation
The company gathers first-person video using a modular camera setup and a uniform collection process with partners. Its software then labels every clip using models like Memories-H0 for hand poses and EgoAction for action segments. The platform also utilizes NVIDIA Cosmos to help interpret incoming video data. This setup allows the team to curate and post-train models tailored for specific industrial and robotics applications.
Building for Physical AI
Memories.ai plans to offer both exclusive and non-exclusive arrangements to support diverse tasks and camera views. Co-founder and CEO Shawn Shen states that the goal is to accelerate physical intelligence and grow alongside the robotics industry. The company has shared elements of its research on GitHub and at major academic events.
How does this story land?
One tap per reader.
Quick poll
Can video data fast-track robot training?
One tap. No sign-up.
Ask this story
Still have a question?
Answers use only what is in this story. They are written by AI and can be wrong.