NXP's BlueBox promises off-the-shelf autonomous cars

An off-the-shelf autonomous car brain could accelerate self-driving vehicles onto the roads, with NXP pitching its open-source based BlueBox to automakers. The platform – which consists not only of the AI smarts for the car but its "nervous system" of sensors and communication – could be used to quickly develop a Level 4 autonomous vehicle, the company claims.

Level 4 is generally acknowledged as the point where an intelligent car could drive itself for a complete journey from beginning to end, with or without human occupants.

So far, no manufacturer offers a production vehicle with Level 4 abilities. The closest to that point, arguably, is Tesla, the Autopilot of which on the Model S and Model X is able to track lanes and preceding traffic, handle braking and acceleration, and switch lanes with a tap of the indicator, though only in "safe" highway conditions.

Tesla's system falls somewhere between Level 2 and Level 3 on the SAE International Standard J3016 scale.

NXP's approach is a full suite of processing and sensing technologies. The BlueBox brain consists of the company's LS2088A embedded compute processor, which has eight 64-bit ARM Cortex-A72 cores each running at 2GHz, in addition to DDR4 memory controllers and automotive-specific accelerators.

It's fanless, delivering its 90,000 DMIPS (million instructions per second) without any need for active cooling.

Paired with that is the NXP S32V automotive vision processor, which includes the GPU, dedicated image processing accelerators, and the ability to draw together input from radar, lidar, and vision sensing embedded around the vehicle.

Finally, there's a second V2X (vehicle-to-X) communication system, which could allow the car to communicate both with other vehicles and with road and city infrastructure.

In total, NXP says, it's enough to give the smart car a 360-degree view around itself at all times. That's fed into the companion software, with embedded machine learning to spot and classify other objects – whether they be cars, pedestrians, road furniture, or something else – along with localizing the car in three-dimensional space, and finally make mapping and driving decisions accordingly.

The BlueBox engine itself is an open platform, based on Linux, designed to be readily customized by automakers. That way, even if the core silicon is the same, they can add their own features and to better distinguish their vehicles from those of rivals.

NXP isn't the only chipmaker looking to score big as cars get more intelligent. NVIDIA, for instance, has chips providing the self-driving smarts of the Roborace autonomous race cars, a real-world proof of concept of the company's Drive PX 2 platform.