MIT researchers reported in a paper dated Nov. 5, 2026, a memory system intended to help robots spot and correct descriptions of objects they may have mislabeled. Called UQ-DAAAM, the method compares several descriptions of the same object seen from different angles and estimates how uncertain the robot is about what it observed. When uncertainty is high, it selects further views to refine the description.

In their comparison, the researchers said the refinement resolved ambiguities for about 92 percent of objects in the robot's environment using only two additional captions per object, and raised object-description accuracy by 5 percent over the state of the art. They said the system also gave more accurate answers about objects and their locations than baseline methods. The work is to be presented at the Conference on Neural Information Processing Systems.

The researchers said next steps include testing UQ-DAAAM with mobile robotic manipulators and applying its uncertainty methods in other settings, while also studying whether high-uncertainty areas could be revisited. The research was supported in part by the U.S. Army Research Laboratory and the U.S. Office of Naval Research.