Researchers at the Massachusetts Institute of Technology have developed a novel microchip capable of generating real-time three-dimensional occupancy maps with a power consumption of merely six milliwatts. Designed to overcome the severe computational and energy constraints of micro-robots, drones, and lightweight augmented reality devices, the chip addresses a fundamental navigation challenge: how to continuously map complex environments while avoiding obstacles without draining limited power reserves. The breakthrough centers on Gleanmer, a dedicated hardware accelerator that translates the team’s previously developed GMMap algorithm into silicon. Traditional voxel-based mapping requires massive memory overhead as resolution increases. GMMap replaces these dense cubes with adjustable Gaussian ellipsoids, compactly representing occupied and free spaces. While the software approach reduced memory demands, running it on portable hardware still exceeded two watts. Gleanmer bridges this gap through hardware-level optimizations that streamline data flow and minimize redundant calculations. Instead of repeatedly processing raw depth images, the chip employs a streaming architecture that converts each frame into a Gaussian representation in a single pass, discarding original data immediately to conserve memory. The design further reduces energy usage through three key innovations. It derives navigable free space directly from compiled obstacle ellipsoids rather than recalculating individual sensor rays, cutting map-building energy by up to sixty-three percent. It clusters spatial queries for route planning, processing adjacent coordinates simultaneously, which boosts throughput by four to ten times while reducing access energy by more than seventy percent. It also implements targeted approximation techniques, lowering numerical precision for less critical parameters and simplifying slope estimations. This reduces the accelerator’s silicon footprint by thirty-eight percent and shrinks final map sizes by up to sixty-three percent without compromising accuracy. Measuring only four square millimeters and equipped with 622 kilobytes of on-chip memory, Gleanmer processes standard 640 by 480 depth images at frame rates between eighty-eight and three hundred thirty-one hertz. It performs half a million to over one million spatial queries per second while maintaining mapping accuracy between ninety-six and ninety-nine percent. In comparative testing, the chip consumes at least 341 times less power than an NVIDIA Jetson TX2 and 44 times less than previous specialized mapping accelerators. This efficiency breakthrough removes a major bottleneck for miniaturized autonomous systems. Micro-robots can now navigate narrow industrial pipelines and traverse cluttered interiors without tethered power. Compact drones and lightweight AR headsets can sustain continuous spatial awareness for extended periods, enabling applications in precision assembly and emergency response. While Gleanmer specifically targets the mapping and querying phase, its integration into complete robotic stacks will require coordination with sensors, localization modules, and control systems. Nevertheless, the chip establishes a new benchmark for ultra-low-power spatial computation, paving the way for autonomous devices capable of operating independently in confined environments.