London, June 18 (SANA) Researchers at the Massachusetts Institute of Technology (MIT) have developed a new framework that enables robots to remember locations, objects and events encountered in real-world environments, a step that could improve how autonomous systems interact with their surroundings.
The system, called DAAAM — short for “Describe Anything Anywhere Anytime” — creates a long-term spatiotemporal memory that allows robots to store and retrieve information through natural-language queries.
According to the research team, the framework combines three-dimensional mapping, computer vision and large language models, enabling robots to link objects not only to specific locations but also to the times at which they were observed.
The researchers said the approach expands on conventional mapping systems, which typically focus on the geometric structure of an environment, by incorporating temporal and contextual information into the robot’s memory.
As a result, robots can retrieve more detailed information and answer practical questions, such as where a particular tool was last used or when a specific object was seen.
The system is also designed to improve efficiency by grouping similar data and selecting representative frames, reducing processing requirements and enabling real-time operation in large and complex environments.
Researchers said the framework uses a language model to interpret user requests and direct searches within the robot’s memory, helping improve response accuracy while reducing errors linked to indirect inference.
The team said the technology could contribute to the development of robots better suited to human-centered environments, with potential applications in manufacturing, service industries and augmented reality systems through the more effective integration of spatial, temporal and contextual information.
R.D/ABD
