Jad Yammine, Malek Wahidi, Abrar Anwar, Yan Chang, Mihir Acharya, Anthony Rizk
IndustrialReMEmbR enables scalable spatiotemporal memory for robot fleets by offloading computation to edge devices and using cached vision tokens for on-demand decoding. It avoids storing raw video and generates scene descriptions only when queried, reducing server load. The system supports GDPR compliance by zeroing out people in image patches during encoding.
It enables large-scale, privacy-preserving monitoring of warehouse operations without overwhelming computational or storage resources. This allows fleet-wide analysis of recurring issues like aisle blockages without continuous high-bandwidth data transfer. The approach is practical for real-world deployment in industrial automation settings.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.