Event

Shared spatiotemporal memory system proposed for robot swarms

1 reports · 1 sources

Jad Yammine and other researchers propose Industrial ReMEmbR, a shared spatiotemporal memory system for warehouse robot swarms that extends NVIDIA ReMEmbR to multi-robot scenarios and reduces server-side computation by generating scene descriptions only when needed. The NVIDIA Jetson on each robot uses Cosmos-Embed1 to encode retrieval embeddings, Cosmos3-Edge to encode visual tokens, and the Wan2.2 VAE to encode reconstruction latents, without uploading raw camera frames. The encoding models are also reused to detect people in image patches and zero them out, supporting personal-information abstraction for GDPR-compliant deployment. The researchers split Cosmos3-Edge at the vision-language boundary, placing the vision encoder on the robot and the language decoder on a local server. This design eliminates caption generation at write time, and the authors state that across all 450 test decodes, the output of cached-token decoding was consistent with the unsplit model.