A research team at MIT has introduced DAAAM (Describe Anything, Anywhere, at Any Moment), an artificial intelligence system that enables robots to form and access long-term memory in real time. The system tracks what a robot sees, where it saw it, and when, allowing it to answer natural language questions about its environment. Using a depth-sensing camera, DAAAM builds a continuously updated “4D scene graph” that records objects, their 3D positions, and timestamps. Tested on large-scale benchmarks, the system outperformed existing methods in accuracy, timing, and navigation tasks.
DAAAM’s key innovation lies in batching visual data for efficient processing, achieving roughly tenfold speed improvements over standard approaches. The system’s Describe Anything Model generates textual descriptions of objects, though it sometimes misidentifies unusual items due to limited training data. Researchers note that while DAAAM operates effectively for ground robots, it may be too slow for drones or virtual reality applications. The team plans to release the code and data as open-source resources.
The project, presented at the Conference on Computer Vision and Pattern Recognition, was supported by the ARL DCIST and ONR RAPID programs. Its developers see potential for use in hospitals, warehouses, and other dynamic environments requiring spatial memory.
Using DAAAM, a robot can quickly access their memory to answer complex queries about its environment in plain language. Here, to answer a query, the robot searches its memory using the word "sculpture" to recall artworks it saw on campus.â
The ‘1 Nojor’ media platform is now live in beta, inviting users to explore and provide feedback as we continue to refine the experience.
