Blog

Are autonomous vehicles the next frontier for the SoC?

New platforms, new requirements, and rules will challenge the suppliers Do you have a GPU in your boat? Your car? Your airplane, tractor, or your motorcycle? Maybe not today, but developments in all those categories are underway. Automobiles, such as the Nvidia powered Tesla S have captured the industry’s attention but they are far from the only platform in development. ...

Robert Dow

New platforms, new requirements, and rules will challenge the suppliers Do you have a GPU in your boat? Your car? Your airplane, tractor, or your motorcycle? Maybe not today, but developments in all those categories are underway.

Automobiles, such as the Nvidia powered Tesla S have captured the industry’s attention but they are far from the only platform in development. However, the GPUs aren’t necessarily being used to drive displays. In fact, it may be a bit misleading to limit the exposition to GPUs—SoCs would be a better model.

Vehicles, like the ones listed above, will used everything a modern SoCs has to offer, sensor input, radio connections, SIMD, GPU, RISC/CISC, DSP, and specialized processors (like ISPs and various state-machines). Why? The first thing that comes to mind is autonomous vehicles. But don’t limit your thinking to automobiles.

Yamaha is developing a motorcycle riding humanoid robot that it hopes will match MotoGP world champion Valentino The Doctor’ Rossi for speed and skill around the racetrack (why it has a humanoid-looking rider is not clear to me).

Drones too are autonomous vehicles, and so are some of the submarines being planned. The U.S. Navy will be launching the autonomous submarine drone named the Slocum Glider. The prototype was tested in 2014, and is currently considering submissions for the Seafox Minehunting and Autonomous Underwater Vehicle (AUV).

Trucks will ply the highways autonomously, only waking up the driver when it’s time to maneuver into a distribution center. Or maybe there will be no driver and they will pull into a yard to then be piloted by a driver to the warehouse. Tracked vehicles in airports and under the streets of San Francisco run all day and night without a driver.

The airborne, smaller (than a Predator) drones with four to six props, are the ones that will use the SoC’s to the fullest. If you haven’t seen the classic work being done at the university of Pennsylvania, check out this page. https:// www.youtube.com/user/KumarLabPenn.

SoCs designed for mobile phones and game machines are being adapted for automobiles, multi-rotor copters, and other autonomous vehicles like vacuum cleaners, lawn mowers, and trash can take and pick up robots.

Those drones, or robots or whatever is your favorite name for these things have gryos, GPS, cameras, accelerometers, magnetometers, barometric sensors, and radios. They all untethered, battery powered and consume prodigious amounts of sensor data. They process that data in real time and make instantaneous decisions about threats, mission, and environment—many of them carry a map in memory to aid them.. Those drones might be bringing that new pair of red shoes you ordered, or following you and recording you falling down a steep hiking trail you should have avoided.

GPUs will be one of the major processing engines in these small, consumer, and corporate vehicles. But the GPUs will have to be expanded to be more useful. The SIMD construction that is so beneficial to driving a computer or smartphone screen needs to be more of a hybrid MIMD/SIMD design, and therein lies the snag. Whereas it would be delightful for the SoC builders to simply re-purpose their existing parts and just put out a development platform and some code examples, to be truly successful they are going to have to develop a specialized version for autonomous vehicles. But the market may be there to justify a special version of the SoC. If you believe the market forecasts for drones (of which there are five and counting) it’s going to be somewhere between a $1 and $8 billion market by 2018, or thereabouts.

The other issue is for some of the markets like the military, automobiles, trucks, and buses, is there will be ruggedness requirements, and regulatory tracking issues. All of a sudden that $15 SoC you thought was going to make you rich in the autonomous vehicle market now has to sell for $50 to cover the overhead issues, and maybe you’ll be able to scale the profit accordingly.

The good news, the fun news, is this is going to be really exciting stuff. These new lonely vehicles are going to demand, and get, the best our industry has to offer in terms of clever algorithms, power management, and compute-density. This emerging market will be much more rewarding than the over hyped VR market, and much more challenging. Will the GPUs be up to it?