Researchers at MIT have built a chip that lets small robots build 3D maps of confined spaces using a fraction of the power current systems require.
The system, named Gleanmer, could allow small autonomous robots to navigate tight, complex environments such as ductwork or HVAC systems, checking for gas leaks without colliding with obstacles.
The chip’s low power draw could also suit lightweight augmented reality headsets used for extended periods on site, including for repair and assembly work.
Generating detailed 3D maps typically demands high power and large amounts of memory, as systems must store camera images and repeatedly process every 3D pixel, or voxel, within them.
The MIT team instead represents obstacles using ellipsoid shapes called Gaussians, which adapt their size and shape to match curved surfaces more efficiently than rigid, cube-shaped voxels. A single elongated ellipsoid can capture a region that would otherwise need many voxels, cutting memory demands substantially.
The chip runs an algorithm called GMMap, developed by the same lab, which generates accurate Gaussians from a single pass of a depth image before discarding it, meaning the system never needs to store a full image at once.
Rather than comparing every pixel to every other pixel, the algorithm assumes neighbouring pixels belong to the same Gaussian, comparing each pixel only to those nearby.
Peter Zhi Xuan Li, co-lead author and MIT graduate student, 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.”
As a robot moves through a space, it often views the same object from multiple angles, generating overlapping Gaussians that would otherwise bloat the map. The researchers developed a technique to fuse these overlapping shapes directly, without returning to the original pixel data, keeping memory and power use low throughout.
Zih-Sing Fu, co-lead author and MIT graduate student, said: “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.”
In testing, Gleanmer reconstructed a range of existing 3D environments and live data streamed from an iPhone camera, generating detailed maps in real time while consuming around six milliwatts of power – around 2.5% of the power required by the best existing mapping chip.
When reusing compact Gaussians to plan a route, the chip allowed a robot to chart a safe path using roughly 20% of the energy otherwise needed.
Vivienne Sze, professor in the department of electrical engineering and computer science 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.
“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. Our chip allows you to store very large maps in a very small space, and do it in a very energy efficient manner.”
The team plans to further improve efficiency by moving processing units closer to the sensors gathering environmental data, and is also exploring whether Gaussians could help AI systems interpret complex blueprints and schematics more efficiently.
Read next: NBS and BDP partner on carbon data platform
Are you a building professional? Sign up for a FREE MEMBERSHIP to upload news stories, post job vacancies, and connect with colleagues on our secure social feed.
