3D occupancy maps divide space into regions that are occupied, free, or unexplored, helping robots understand their surroundings. But building high‑fidelity maps usually demands a lot of power, which limits small, battery‑driven devices.
A team of researchers at MIT has now created a system‑on‑a‑chip called Gleanmer that lets small autonomous robots create detailed 3D maps in real time. Imagine drones zipping through narrow HVAC ducts to check for gas leaks; Gleanmer makes that possible.
The chip is incredibly energy‑efficient, consuming just 6 milliwatts of power, about the same as a single LED light, and only a fraction of what other mapping systems require.
Building 3D maps usually demands heavy computing power and large memory to store the environment. The MIT team solved this by pairing an ultra‑efficient mapping algorithm with custom hardware designed to speed up the workload. This clever combination slashes both memory use and power consumption, making it possible for tiny, battery‑limited robots to generate detailed maps in real time.
MIT’s Clio enables robots to quickly map task-relevant objects
“While there has been a lot of work looking into compact 3D maps,” says Vivienne Sze, senior author of a paper on the chip, “what stands out about this work is that it also ensures that the process to generate those maps is as efficient as possible. Our chip allows you to store very large maps in a very small space, and do it in a very energy-efficient manner.”
Gleanmer is powered by a custom accelerator for GMMap, a 3D occupancy mapping technique that represents obstacles as ellipsoid “blobs” called Gaussians. The chip can generate these Gaussians directly from depth images in a single pass, then discard the images, meaning it never has to store full pictures at once.
Gleanmer attains remarkable efficiency as a result of closely co-optimizing its algorithm with the hardware: 63% less energy for map construction; 81% less energy for map queries; 38% smaller accelerator area thanks to approximate Gaussian computation.
Built on 16nm CMOS technology, Gleanmer processes 640×480 images in real time at over 88 frames per second and handles more than 540,000 coordinates per second during map queries.
As researchers noted, “To our knowledge, Gleanmer is the first fabricated SoC to achieve real-time 3D occupancy mapping under 6 mW for edge-based applications.”
The researchers want to make the chip even more energy‑efficient by moving its processing units closer to the sensors that collect environmental data. They’re also exploring new uses, like representing schematics with Gaussians, which could help AI systems understand and work with complex blueprints more effectively.
Journal Reference:
- Zih-Sing Fu, Peter Zhi Xuan Li, Sertac Karaman, Vivienne Sze et al. Gleanmer: A 6 mW SoC for Real-Time 3D Gaussian Occupancy Mapping.