Autonomous startups are exploding right now and the latest one to catch my attention is called Ghost. The company has divisions in both Mountain View California and Sydney Australia and started back in 2015.
Aiming to deliver Level 3 autonomy to existing vehicles, Ghost says their solution is compatible with most cars since 2012. That should excite existing vehicle owners who want smarter vehicles, but aren’t keen on buying a whole new car to get it.
The hardware has multiple components and to integrate with the car, requires a professional installation which will cost US$3,495. On top of this you’ll need to pay US$99 per month to use the system, certainly not cheap, but for that you do get some big promises.
Chost’s website claims you get
- Real Collision Avoidance
- Superhuman Reflexes
- No Oversight Needed
- No Bugs, No Glitches
That’s an incredibly ambitious statement from a company that is yet to ship any products and doesn’t make their own vehicles. We’ve seen many companies promise a big game when it comes to driverless vehicles, so naturally, I’m skeptical. Lets take a look at their technology to see if it offers and hints about why they’re so confident.
Ghost says their developing a collision-avoidance system that can handle any obstacle, no matter what it is, in any size, shape or lighting condition. To do this, they’re using the same processing functions as the human visual cortex, combining high-definition perception and relative motion to detect any potential obstacle in your path to keep you safe.
To break that down, Ghost is using Machine Learning AI, specifically Computer Vision, which places its retrofit solution in competition with Comma.AI and the market leader in that space, Tesla.
The massive disclaimer here is that ‘Real Self-Driving’ by Ghost, only works on the freeway – Ghost takes the wheel on the freeway so you can do something else entirely. That may change in the future, but it certainly limits the value proposition.
Ghost use a tech stack that includes:
- 8x 13 MP cameras
- 8x 120˚ HD wide-angle lenses
- Pixel binning for low-light sensitivity
- 30 fps
Self-driving computer (located in the boot)
- 40x CPU cores (5x 8-core mobile chips)
- 5x Graphics Processing Units
- 640 GB of Flash Storage
Mounted to the windshield
When it comes to safety, Ghost claim they can enhance a vehicle’s capabilities to by super-human, seeing the road better and reacting 3x faster than a person.
No oversight needed
Here’s perhaps the boldest claim of all. Ghost says their technology is capable of driving without driver intervention.
Instead of relying on the driver to intervene at a moment’s notice, Ghost’s Real Self-Driving system is designed with fallbacks at multiple levels so the car can continue to drive safely even in the case of unexpected driving scenarios or mechanical failures.
Every level of the system has a backup to maintain uninterrupted driving, from multiple cameras, computers and cables in hardware to additional layers in the runtime, perception and driving models of the software stack.
There’s no mention of driver monitoring systems, but it sounds like Ghost’s technology aims to provide advanced warning to the driver should they need to take back control when it encounters edge cases. The goal here is to move past what we have with Level 2 systems that require drivers to be prepared to take control at any stage and actually relax and enjoy the ride.
While I’d love more competition in the autonomy space, it is hardware to know the potential chances for success of Ghost until we see their technology in action. When using ML you essentially attempt to reproduce human ability, giving the car a vast array of driving knowledge and decision making capabilities like drivable space, lanes obstacle avoidance etc. How Ghost will acquire the real-world data required to build a hands-off, level 3 system is the big question.
Ghost is currently accepting emails to notify you when the hardware and software are available.
Read more at https://driveghost.com