Ghost Autonomy Wants to Use Smartphone Cameras to Crack Autonomous Driving

As a child, we would “ghost ride” our bikes. It was a destructive endeavor where we’d jump off our bicycles – typically ahead of a ramp – and watch the ensuing carnage with glee. Decades later, ‘ghost riding the whip’ became a short-lived phenomenon where a driver would get out of a car as it rolled down the road and dance alongside the rolling hunk of metal and glass before (hopefully) jumping back into the driver’s seat. It was YouTube fodder that resulted in far too many collisions.

So it’s interesting that Ghost Autonomy would pick, well… that name. Yes, it conjures images of apparitions piloting vehicles while the driver relaxes and possibly watches old videos of people dancing near cars. Who wouldn’t want a friendly ghost driving them around? But as I enter the company’s headquarters in Silicon Valley all I hear is E-40’s Tell Me When to Go (now scrape, scrape, scrape, scrape) in my head.

The reality is that Ghost is trying to democratize autonomous driving. While there are no driverless vehicles for sale in the United States right now, the goal is that eventually, people will be able to let go of the wheel and relax while their vehicles deliver them to a destination. The issue is that it’s a very hard problem to solve. Tesla has been promising the feature for years and the automotive startup world is littered with the corpses of companies that thought they could crack the code while burning through millions of investor dollars to bring autonomy to the masses.

Beyond the burn rates of startups, it’s also very expensive. Pricey sensors and high-end computers put the feature out of reach of most vehicles and self-driving will likely make its debut in cars and SUVs that cost around $100,000 or more. Ghost Autonomy wants to deliver autonomy to $30,000 vehicles.

Cell Phone Camera Sensors

Ghost believes it can accomplish this by standing on the shoulders of larger companies. Mainly the corporations that build the high-end cameras in your cell phone, which currently have far more resolution than what many automakers are putting on their vehicles. Ghost places a longer lens so the vehicle can see further and now you have a high-definition sensor on your vehicle.

Ghost Autonomy CEO John Hayes notes that Ghost and really no LIDAR or radar company can match the billions of dollars that go into the research and development of the cameras in our phones.

Plus, they’re cheap. A camera cost Ghost about $12 a pop. The CPU that runs all of this, it’s about $45.

“The only thing that people care about in the entire cell phone industry is how did you make the camera better,” Hayes said. The CEO added that cameras are,” improving at a rate that’s faster than the rate that LIDAR is improving.”

Plus, they’re cheap. A camera cost Ghost about $12 a pop. The CPU that runs all of this, it’s about $45. The entire sensor package with multiple cameras creates a stereoscopic view of the world 360 degrees around the vehicle. Two cameras pointed in every direction gives the system depth perception the way the two eyes in your head let you determine how far away objects are.

Ghost is also using radar, but not LIDAR. That’s not to say the company is anti-LIDAR like Tesla. Instead, it’s just too expensive right now. The goal is to get the entire camera and CPU suite to under $1,500. But really the company is not in the business of building hardware. It’s merely showing what can be done with less expensive, off-the-shelf hardware so that tier-one suppliers can build the gear for automakers. Instead, it’s banking on creating the software that uses that inexpensive sensor suite.

All Objects Are Objects

Like its hardware strategy, Ghost is also doing something a bit different with its Ghost OS software. Instead of training its AI to recognize everything in the world, it just wants the system to know that something is there.

For example, instead of teaching the system to identify things like dogs, children, trees, cars, and bikes, the system just knows when it sees an object, where that thing resides in space, and if it’s moving or not, and then classifies it based on where it is and how quickly it’s moving. For example, an object that’s roughly the shape of a human and moves at the speed of a human is probably a human and the vehicle needs to give that person a bit more space because it’s a vulnerable road user.

“The base fundamental algorithm doesn’t require identification to understand what’s free space and what’s not and that allows us to not have the risk of misidentification,” Ghost Autonomy’s vice-president of marketing and design, Matt Kixmoeller told Motor1.com.

Ride Along

Typically any driver assistance function requires a button to be pressed. Usually, it’s to set the adaptive cruise control and maybe there is a second control to enable the lane centering. The Ghost system is built to work with the driver in what Hayes calls “collaborative AI.”

As you drive the system turns itself on and helps out. You move your foot from the accelerator or your hands from the wheel and Ghost takes over. The goal is for your car to drive essentially the way you do and to keep up with traffic without having to enable anything. It’s just on and ready to go when you are.

For example, if everyone is doing 75 miles an hour when the speed limit is 65, the Ghost-enabled car will do 75. Instead of sticking to the center of the lane while passing a large truck, it’ll move over a bit. It’ll even move slightly to the inside of a curve.

The system is still in the testing phase so legally I was unable to get behind the wheel, but what I saw was impressive.

One thing that’s an issue with driver assistance systems is mode confusion, it’s why Mercedes, General Motors, and BMW have lights on their steering wheels for their hands-free driver assistance features. It’s to signify what mode the vehicle is in at any moment so the driver knows when they can safely take their hands off the wheel.

Ghost’s driver assistance doesn’t have that. Instead, it’s there with you the whole time while on the highway and understands when you’ve taken your hand off the wheel. When you’re ready to take back over, you place your hands back on the wheel and you’re in control with having to wrestle control away from the system.

Currently, the Ghost OS has a basic highway pilot function and supports automated lane changes. It’s doing all this in a Toyota Camry test vehicle the company has specially outfitted with hardware.

The Business Of Giving Everyone Autonomy

A Camry is pretty far from say a Mercedes S-Class which will likely be the first Level 3 vehicle to go on sale in the United States. Ghost’s inexpensive hardware solution means inexpensive cars (like a Camry) cars can be outfitted with self-driving hardware and drivers can subscribe to turn on Ghost OS.

Automakers are eyeing subscription models and while the potential buyers of luxury vehicles can afford expensive autonomous systems, those buying a Camry would likely be priced out if an automaker added the thousands of dollars in hardware and software from a level 3-capable S-Class.

But, if an automaker can throw a $1,500 (or less) suite of hardware on all cars and then get folks to pay a little extra every month for the chance to drive without actually driving, Ghost could become very popular with OEMs. The company says it’s already talking to automakers but of course, can’t say which ones. What Ghost can say is that it expects something to be on the road as a 2025 model.

So you could soon ghost-ride your not-too-expensive whip without the dancing or YouTube videos that end in disaster. Or at least, that’s the plan.