Gleanmer

Gleanmer is a research chip developed at the Massachusetts Institute of Technology's Research Laboratory of Electronics (RLE), not a commercial company, presented as a low-power system-on-chip that lets small, battery-powered robots build detailed 3D maps of their surroundings in real time and plan collision-free paths through them. The work was led by Vivienne Sze, an MIT professor of electrical engineering and computer science, with Peter Zhi Xuan Li and Zih-Sing Fu serving as co-lead student authors. The chip consumes approximately 6 milliwatts of power during operation, which the research team describes as roughly 2.5 percent of the energy required by the best previously existing mapping-specific chip designs, and comparable to the draw of a single LED. Instead of representing an environment with millions of small cubic voxels, as conventional 3D mapping systems do, Gleanmer implements the GMMap algorithm, which models scenes using flexible ellipsoidal shapes called Gaussians whose size, shape, and orientation adapt to fit surfaces and open spaces. The chip generates these Gaussian representations directly from depth-camera images in a single processing pass and then discards the original image data, and it fuses overlapping Gaussians observed from different viewpoints without needing to revisit the original pixel data, substantially cutting memory and power requirements compared to voxel- or point-cloud-based approaches. Gleanmer keeps recently observed scene data in fast on-chip memory rather than relying on power-hungry off-chip DRAM, and the research team reports the chip can process 640-by-480-pixel depth images at more than 88 frames per second, answer over 540,000 spatial map queries per second, and reduce the energy needed for trajectory/path planning by about 20 percent relative to conventional methods; it is fabricated on a 16-nanometer manufacturing process. Target applications described by the research team include autonomous navigation and obstacle avoidance for miniature drones and other small robots operating in confined or hard-to-access spaces such as industrial HVAC ductwork, warehouses, and pipelines, as well as lightweight augmented-reality headsets and exploratory use in interpreting technical drawings and schematics. The research was supported in part by an MIT-MathWorks Fellowship, Amazon, the U.S. National Science Foundation, and Intel, and was presented at the IEEE Symposium on VLSI Technology and Circuits, with a preprint also posted to arXiv. Because Gleanmer is an academic research prototype rather than a commercial product, there is no company headquarters address, product pricing, sales contact, or corporate social media presence to report; the project is documented on MIT RLE's website and in MIT-affiliated technical press coverage.