Progress? Autonomous Driving, SDVs, and Artificial Intelligence Updates From CES 2024

Listening for these buzzwords in a CES 2024 drinking game would have left players well and truly schnockered.

ManufacturerPhotographerJan 16, 2024

CES 2024 has been a hotbed of novel sensor-technologies and improvements on existing tech, and this year was no exception with two companies boasting perception solutions that work like eyeglasses for your car—improving vision from 20/20 to 20/10 (by doubling a sensor's range). One of these was hardware-based, the other software/math-based. We also took a spin in a fully autonomous vehicle that managed the drive about as well as a teenager taking the wheel for the first time. Also, several interview subjects weighed in on the question, "Is it better for an OEM to develop all software in-house, or buy it from suppliers?" Finally, we uncovered a regulation that could be a real buzz-kill for all those over-the-air updated features we love getting on our SDVs. Here's everything we learned and can expect going forward as automotive technology continues to rapidly advance in the age of the software-defined vehicle.

Neural Propulsion Systems—Sooner, Clearer, Farther

We covered the Neural Propulsion System multi-modal radar setup and explained how it can see around corners using math from its AtomicSense platform to condition the info coming back from the radar emitters. See, radar is awesome in fog, rain, etc.—way better than Lidar—but it can't generate the lifelike point-clouds that Lidar does and it tends to struggle mightily with "ghost images," which are reflections from radar waves that may have bounced off of other things, giving a false positive that could result in an annoying and utterly unnecessary automatic emergency braking event.

NPS' math and software runs on standard radar chip sets from TI Automotive or others, boasting resolution better than 10x, suppression of false positives in the 10x realm, to reliably detect objects twice as far away as the standard chip could in part by measuring the Doppler shift and the range simultaneously, where others typically do this sequentially. This can result in greater safety or lower cost (by cutting the number of radar sensors in half, if desired). NPS claims it can achieve 0.5-degree angular resolution with just two chips, whereas TI requires four and Mobileye reportedly needs 14 for that resolution. A major OEM will be announcing investment soon, and the company has forged a new relationship with defense radar supplier Raytheon.

Provizio 5D Perception

Don't roll your eyes at "5D" so quick! Provizio reps are quick to note that it's marketing speak, with the fifth "dimension" being perception/object detection and identification happening on the chip. Here again, Provizio is using bulk radar chips, but in this case it arranges them slightly differently so the multiple-input/multiple-sparse-output (MIMSO) antenna design is bespoke and results in 6,032 virtual apertures.

This design promises detection to as much as 600 meters, and the company claims its product makes accurate enough point clouds to replace lidar. The system can also map indoor spaces where there's no GPS signal. Provizio introduced 5D Perception at CES 2024, so there are no buyers lined up as yet, if it really can replicate the strengths of Lidar at a radar price point, Provizio is likely to find plenty of takers.

Ambarella Autonomous Drive

Chip-manufacturer Ambarella undertook an ambitious trip from Italy to Shanghai, China in an autonomous vehicle in 2010. That vehicle required 18 PCs in the trunk to run all the perception technology on the car. Flash forward 14 years, and we were able to take a ride in a Lexus equipped with just one small computer on the rear cargo deck running the company's CV3 chips, taking signals from an Oculii radar unit in front plus four corner radars and 18 cameras arrayed around the vehicle.

A simple low-definition map indicating only the centerline of every road is used, with the vehicle using neural-net processing to fill in all the HD perception required, thanks to machine learning, neural networks, and artificial intelligence (AI) tech. Ambarella even demonstrated an AI-based large-language model to help identify peculiar objects on the road (an example: a three-wheel bike grossly overloaded with cargo being driven by a man smoking a cigarette—and all of these details showed up in the LLM description generated in an instant).

Last year at CES, this vehicle only demonstrated the perception, but this year it drove itself around—with mixed results. The system needs a lot of work on its acceleration and braking smoothness, and it nearly hit a curb once before the safety driver saved the day. Still, it's an impressive leap in capability from last year.

OEM Vehicle Software: DIY or Buy?

Appearing on the Fully Charged Podcast in August, Ford CEO Jim Farley lamented the state of his company's software development, noting that the many subsystems comprising a modern FoMoCo product may involve 150 different pieces of software developed by as many different suppliers—each of which must be consulted to implement revisions. "That's why at Ford we've decided in the second-generation [EV] product to completely insource electric architecture," he announced.

Jan Becker, CEO of Apex.AI, 2024MotorTrendSDV Innovator Awards Pioneer, and one of the three authors of the original document spelling out the five levels of autonomous driving capability, lauded the goal while cautioning that may not be the most resource-efficient approach. He noted that Tesla and some Chinese startups were able to in-source their entire software stack because no legacy architecture existed. Legacy manufacturers like Ford, Becker reckons, will waste time attempting to start from scratch. Smarter to employ existing open-source software-development kits (SDKs) that can affect an "abstraction layer" between the hardware and the user interface. Apex.AI offers just such a middleware solution in Apex.OS, which leverages open-source software platforms like ROS or Eclipse iceoryx, adding to them whatever is needed for safety-critical vehicle systems. And somewhat uniquely, Apex.AI hands over whatever source code it develops to the OEM.

Throughout the week, Tier 1 suppliers repeatedly told us they were prepared to provide turn-key software or work with manufacturers to adapt existing or new software. This will be an interesting space to watch.

UNECE R155 "CSMS" Could Kill SDV Features

The United Nations Economic Commission for Europe Regulation 155 requires vehicle manufacturers to provide a cyber security management system (CSMS), and to prove that it works to gain type approval required to legally sell and register a vehicle. A corollary regulation, R156 stipulates that for attacks that become known, updates must be written and pushed out to the vehicle fleet via a software update management system.

Derek de Bono, Valeo group software defined vehicle product vice president, expressed concern that some of the beloved features built into or updated onto our vehicles may end up needing to be over-the-air removed if, for example, the computing power of the built-in hardware becomes insufficient to run required cyber-security upgrades to protect the vehicle.

Argus vDome to the Rescue?

Cybersecurity experts Argus showed off two potential cybersecurity solutions, to detect and thwart attacks on a vehicle's electrical architecture. There's a software solution for hacks coming in over the internet, and OEMs can specify whether to be alerted to every attack on one of their vehicles (probably a good idea). The other is a hardware solution, dubbed vDome. This one protects against CAN bus attacks, using malicious devices that can be obtained on the internet for just one or two hundred dollars. These often connect to a headlamp plug, and from there unlock the doors and start the engine.

The vDome device would have to be developed in conjunction with OEMs to work out where it can be inserted into the CAN bus, but could then be sold in the aftermarket, providing peace of mind and perhaps lower insurance rates.

More From CES 2024:

Frank Markus

I started critiquing cars at age 5 by bumming rides home from church in other parishioners’ new cars. At 16 I started running parts for an Oldsmobile dealership and got hooked on the car biz. Engineering seemed the best way to make a living in it, so with two mechanical engineering degrees I joined Chrysler to work on the Neon, LH cars, and 2nd-gen minivans. Then a friend mentioned an opening for a technical editor at another car magazine, and I did the car-biz equivalent of running off to join the circus. I loved that job too until the phone rang again with what turned out to be an even better opportunity with Motor Trend. It’s nearly impossible to imagine an even better job, but I still answer the phone…