Breaking01 / AI
MORSEPose learns to estimate pose without surface models
The method uses labels for object pose and physical size to select observed points for surface reconstruction during training.

Researchers introduced MORSEPose on 20 Nov 2026 in a paper describing how it estimates the position and orientation of objects it has never seen before.
According to the paper, MORSEPose takes colour and depth images that show only part of the object.
The researchers used labels for the object's pose and physical size to set a boundary with a margin. They used observed points inside that boundary to train the method to reconstruct the surface.
The researchers reported pose estimation performance on REAL275 and HouseCat6D comparable to that of existing methods.
The existing methods in the comparison selected observed points based on their distance from a model of the object's surface.
The paper's abstract did not give pose estimation accuracy figures for either benchmark.