Sensor Fusion Software Simplifies Design of Robots, Autonomous Cars

Using a novel arithmetic approach, a new sensor fusion solution fits on a single microcontroller, allowing developers to use it in applications ranging from autonomous cars to robots to drones.

Known as SigmaFusion, the new software product could help engineers more effectively deal with the blizzard of data coming from radar, Lidar, cameras, and ultrasonic sensors, while simultaneously enabling them to reduce the number of electronic control units in their overall system. “We’ve shrunk down the software to make it available to a wider class of applications,” Julien Mottin, research engineer for the Leti Research Institute of CEA Tech, told Design News. “And there’s no drawback in terms of mathematical precision, even though we can run it on a much smaller platform.”

Mottin said the key to the product’s capabilities is its ability to execute so-called occupancy grids without the need for a floating point unit. Using an integer unit instead of floating point, he said, makes for a smaller software footprint. Because it is less compute-intensive, it can be incorporated in a single MCU, and yet still control powertrain, body-safety, and ADAS applications in vehicles.

“The major problem with those occupancy grids is that in the current state of the art, it’s a challenge to run them in hardware,” Mottin said. “They use complex computation techniques that (typically) rely on floating point arithmetic.”

But by eliminating the need for floating point arithmetic, Leti engineers say their product is a candidate, not only for autonomous cars, but for smaller applications. Mottin said the technology could be applied to small robots and drones. He added that the company is even working with a supplier on the use of SigmaFusion in a cane for the blind. In that application, the sensor fusion package could draw information from sensors and provide an audible warning of potential obstacles, not only in the path of the user’s feet, but in front of his or her head. The sensor fusion software could help keep costs down in such applications by enabling its use without the need for a costly hardware platform.

Mottin added that SigmaFusion also offers the advantage of being able to characterize free space, as well as obstacles, in the path of the sensors. Typically, he said, the occupancy grids that form the backbone of the software are capable only of distinguishing obstacles – such as trees, cars, and pedestrians. But by characterizing free space, he said, it provides another level of assurance regarding the road ahead.

“To fully assess safety, you really need to know, not only about the obstacles, but how much free space you have in front of you,” he told us.

The technology has already been embedded in Infineon Technologies’ Aurix TC29x platform. Leti is also working with other, unnamed, suppliers on incorporation of the sensor fusion package into their MCUs.

Mottin said he expects initial applications to be in the automotive space. “The difference is that we’ve been able to shrink the software down so it fits in automotive-grade platforms, especially those used in ADAS,” he said. “With this, there’s no need for complex graphics processing units or accelerators.”

Senior technical editor Chuck Murray has been writing about technology for 33 years. He joined Design News in 1987, and has covered electronics, automation, fluid power, and auto.