George Hotz is the founder of comma.ai, who is actively integrating 3rd party hardware on existing vehicle ADAS systems. This approach is often referred to as ‘the Android’ of self-driving as it allows a new level of customisation, including the use of driving monitoring, rather than the strict hands-on the wheel requirement.
On September 6th, comma.ai went live on YouTube, streaming a pre-release build of their software. This demonstration showcased openpilot performing lane changes, a key part of the step towards full self-driving.
Their system is progressing quickly, also leveraging loads of data of how people drive, crunched by a neural net, to produce everything from lane centering, path planning and drivable space.
The early build of the software allows the car to determine where the lanes are and transition from one to another, without requiring input from the driver.
What’s not yet built is the ability to check if the lane is clear before making the lane change, for not at least, that’s the responsibility of the driver.
Hotz explains that lane changing is a 0.7 feature, and is the result of data gathered from around 100,000 lane changes done by drivers who have openpilot hardware in their cars already.
Perhaps one of the biggest take aways from the video is the discussion that’s had around the ability to software-control the indicators. Hotz questions if the only way to initiate and disable the indicators is by physically moving the indicator stalk.
The unfortunate reality is that traditional automakers built their cars always expecting a human to be operating it. This means that the dream of having the left or right indicator be activated by openpilot software, is simply not possible. The only other time indicators are activated is through another hardware button, the hazards. While it’d certainly be possible to wire in a solution, the whole aim of comma.io is to enable self-driving by bolting on a couple of quick and easy aftermarket parts.
It is precisely differences like this, that means, as admirable as comma.ai’s autonomous endeavours are, they will never seriously compete with a company like Tesla that has full control over the design and development of every aspect of the vehicle.
Recently Hotz was interviewed by Lex Fridman and discussed in detail the company’s efforts to catch the industry leader in this space, Tesla. It’s an almost 2-hour interview and well worth watching.
The point you’re missing is that the Comma.ai vision is to eventually get their software into the car manufacturing process such that it would be an add-on you buy with the car – at this point this is 100% open source research which is admirable in it of itself, but the vision is much more refined than slapping an ugly component on the windshield + aftermarket car controllers…
Does that solve the problem brought up by the article of not being able to control some important aspects of the car (e.g. turn signals) via CANbus?
just make a device that can pull the blinker’s lever up or down. it would be a simple thing to engineer.