MIT releases UQ-DAAAM robot memory system

MIT researchers have developed a new robotic memory system called UQ-DAAAM, which integrates uncertainty quantification into vision-language models to correct mislabeled or blurry object descriptions. The system compares multiple viewpoints of objects as a robot moves, selects additional frames to reduce uncertainty, and uses an LLM to refine descriptions in real time. This approach improves the accuracy of object detection and memory retention, helping robots safely perform complex tasks. Open article →