CAMBRIDGE — Researchers at the Massachusetts Institute of Technology developed Gleanmer, a new system-on-a-chip designed for energy-efficient real-time 3D mapping. The chip combines an algorithm called GMMap with specialized hardware to generate 3D maps for navigation, consuming approximately 6 milliwatts of power.

The Gleanmer chip represents obstacles and free space in 3D maps using ellipsoids, known as Gaussians, instead of cube-shaped voxels. This GMMap algorithm generates Gaussians from depth images in a single pass, allowing the system to discard images without storing them fully. The algorithm assumes nearby pixels belong to the same Gaussian, comparing each pixel only to its neighbors, and performs fusion of overlapping Gaussians directly.

The chip stores active Gaussians in on-chip memory adjacent to its computational units. Vivienne Sze, a professor at MIT and senior author of the paper, said, "This paper showcases a key example of how you can leverage co-design of the algorithm and hardware to really push energy efficiency." She added, "While there has been a lot of work looking into compact 3D maps, what stands out about this work is that it also ensures that the process to generate those maps is as efficient as possible." Sze also stated, "Our chip allows you to store very large maps in a very small space, and do it in a very energy efficient manner."

Peter Zhi Xuan Li, an MIT graduate student and co-lead author of the paper, said, "At any point in time, we only need to store a few pixels in memory, which significantly reduces the memory footprint our algorithm requires." Zih-Sing Fu, also an MIT graduate student and co-lead author, explained, "By having a dedicated memory that just stores the objects you've seen in the previous few frames, you can access the data much more efficiently."

The researchers tested the chip by reconstructing pre-existing 3D environments and by using live data from an iPhone camera. The work was presented at the IEEE Very Large-Scale Integrated Circuits Symposium. Professor Sertac Karaman, director of LIDS, is also a co-author of the paper.