MIT researchers developed Gleanmer, a system-on-a-chip that enables tiny autonomous robots and low-power devices to build detailed real-time 3D maps of their environments using only ~6 milliwatts of power — about 2.5% of what existing mapping chips require. The key innovation is replacing traditional voxel-based mapping with Gaussian ellipsoids, which represent obstacles and free space far more compactly. A single-pass algorithm processes depth images pixel-by-pixel without storing full frames, and overlapping Gaussians are fused directly without revisiting raw pixels. This co-design of algorithm and hardware keeps active data in fast on-chip memory, slashing energy and memory needs. Potential applications include UAVs navigating tight spaces, AR headsets, and AI systems reasoning over complex blueprints.
