High-efficiency 3D map builder

Researchers in the US have developed a low-power chip for building 3D maps for small robots and autonomous ground vehicles, writes Nick Flaherty.
The Gleanmer chip constructs detailed 3D maps of an environment for path planning without processing multiple pixels. Instead, the researchers at MIT took a different approach by combining an extremely efficient mapping algorithm with specialised hardware designed to accelerate its workload, which minimises memory and power consumption.
Built in a cost-effective 16 nm CMOS process, the chip consumes only about 6 mW – a fraction of the power required by other 3D mapping systems. It processes 640 x 480 pixel images in real time beyond 88 frame/s during map construction and processes over 540 K coordinates per second during map query.
Instead of representing the environment with 3D pixels, or voxels, the technique maps obstacles in space using ellipsoid blobs called Gaussians. The size, shape and thickness of these ellipsoids can be smoothly adapted, so they match the shape of curved objects more efficiently than if one uses rigid, cube-shaped voxels.

Gaussians can flexibly fit the geometry, so a single elongated ellipsoid can represent a region that would take many voxels. This means that occupied surfaces and free space are captured far more compactly.
The chip implements an algorithm called GMMap to generate a 3D map of a robot’s environment using Gaussians in a single pass. The algorithm assumes that nearby pixels belong in the same Gaussian, so it only needs to compare each pixel to its neighbours.
“This is 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,” said researcher Prof Vivienne Sze in the Department of Electrical Engineering and Computer Science at MIT and a member of the Research Laboratory of Electronics.
As the robot moves through the space, it usually sees the same object from different viewpoints. Some Gaussians will overlap because they represent the same object and this can make the 3D map too large to store on an edge device.
Fusing overlapping Gaussians makes the map more compact, but doing so typically requires the algorithm to process many raw pixels stored in memory. The researchers developed a novel technique to perform this fusion process directly on overlapping Gaussians in the on-chip memory without needing to revisit the original pixels. This significantly reduces memory and power requirements.
The researchers plan to further improve energy efficiency by moving the processing units on the chip closer to the sensors that gather the environmental data.
UPCOMING EVENTS