Autopilot already outpacing lexicon

Early this year, TeslaMondo wrote that we’re headed for a tangled mess of terminology as cars become self-aware. Well, now that Tesla’s Autopilot has arrived, the chaos has begun. The news coverage this morning runs the gamut from “hands-free” to “sort of hands-free” to “driverless” to “sort of driverless” to “keep your hands on the wheel” to “you don’t have to touch the wheel.” Meanwhile, the plaintiffs’ bar undoubtedly is on high alert. This terminology is going to become legally significant upon the first accident.

Sigh.

And the most interesting angle of the story — the continuous feedback loop that allows each car to learn from the others —  is the most difficult to grasp and therefore given short shrift by reporters on deadline. HOW EXACTLY DOES THIS WORK? What constitutes a beneficial experience that should be shared with the fleet? Are there negative experiences that should not be shared? If, so, how are they filtered? And how exactly does my car learn from your car, and how fast?

If there’s an article out there that explains this stuff, please link.

Tagged

4 thoughts on “Autopilot already outpacing lexicon

  1. afarnham says:

    I make software for a living, some of it in orbit right now, but I am not a full expert on driverless cars. That said I know how I would tackle the problem.

    The general idea is to have a really good classification system. What you want is to be able to capture every driver override of the system and tag the various pieces of that with a set of well defined actions as well as the current state of the world around the car and the longer term outcome of that override. If it it leads to a crash in the next 5-10 seconds, perhaps the driver was wrong.

    As drivers override things, sets of action, condition, and outcome tags can be grouped together using nearest neighbor algorithms. The groups that emerge can then be classified as good or bad things, probably by humans, and fed to a machine learning algorithm with the appropriate classification of good or bad.

    Like

  2. RWF says:

    This sounds awfully slow and labor-intensive for a smallish company. Human input for every improvement scenario?

    Like

  3. Andy says:

    I think thats how i would do it too, but as an aside, FB has a 265B market cap for being a data company. No other automaker has this type of telematics baked in. When the market realizes the value of the data, then i would expect the stock price to freak out.

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s