Zih-Sing Fu adjusted a tiny drone in MIT’s robotics lab, its rotors humming like a bumblebee, as the new Gleanmer chip inside began rendering a 3D map of the cluttered room in real time—using less power than a nightlight. This breakthrough, developed by MIT researchers, could redefine how miniature robots navigate complex, confined spaces, from detecting gas leaks in industrial HVAC systems to assisting surgeons through augmented reality headsets. For years, the challenge has been energy: building detailed 3D maps demands heavy computation and memory, far beyond what a palm-sized drone or lightweight AR glasses can handle. But the Gleanmer chip changes that equation entirely. By co-designing a novel algorithm with custom hardware, the team has created a system-on-a-chip that consumes just 6 milliwatts—barely a fraction of what traditional mapping systems require. This leap in efficiency opens the door for long-lasting, intelligent micro-robots and wearable tech that can operate autonomously for hours.
The innovation hinges on a smarter way of seeing space. Instead of using rigid 3D cubes called voxels to map obstacles, the MIT team uses smooth, adaptable ellipsoids known as Gaussians. These blobs stretch and reshape to fit objects more naturally, capturing curved walls or pipes with far fewer data points. A single elongated Gaussian can replace dozens of voxels, drastically shrinking the map’s size. The team’s algorithm, GMMap, builds these maps in a single pass through a depth image, comparing only neighboring pixels—eliminating the need to store entire images in memory. As Peter Zhi Xuan Li explains, “At any point in time, we only need to store a few pixels in memory,” slashing both power and storage demands. When the robot moves and sees the same object from another angle, overlapping Gaussians are fused on the chip itself, without revisiting raw pixel data—a first in edge-based mapping.
The implications stretch beyond robotics. In augmented reality, where battery life limits immersive experiences, Gleanmer could enable lightweight headsets to maintain rich, persistent 3D environments for medical training or complex assembly tasks. In hazardous environments, swarms of insect-sized drones could map gas leaks in ventilation systems without risking human inspectors. Led by Vivienne Sze, Sertac Karaman, and their graduate students, the team presented their work at the IEEE Very Large-Scale Integrated Circuits Symposium, where it drew attention for its elegant fusion of algorithm and hardware. “This paper showcases a key example of how you can leverage co-design of the algorithm and hardware to really push energy efficiency,” says Sze. As edge computing grows, chips like Gleanmer may become the quiet engines behind a new generation of intelligent, invisible machines—mapping the world not with brute force, but with brilliance.