Quote:
Originally Posted by KFC911
(Post 12510937)
don't know the difference between Waymo or robotaxi
|
Waymo is a robotaxi service. There are others as well.
Quote:
Originally Posted by KFC911
(Post 12510937)
Are you really locked in a car
|
not sure about all robotaxi services but at least with the one I worked on you are no more "locked in a car" than with a regular taxi or uber or whatever
Quote:
Originally Posted by KFC911
(Post 12510937)
at the mercy of "code"
|
yes, "code" is operating the car
Quote:
Originally Posted by KFC911
(Post 12510937)
and "network connectivity"
|
network connectivity is not really required per se, all of the autonomous capability is within the car itself. There are computers (and redundant computers) within the car doing all of the driving. However, the cars do use GPS (just as you do... to determine the best route etc) and have cellular connectivity so that they can come and pick you up (just like uber) as well as to connect with remote operators who can assist passengers, take over and operate the car if needed. So if suddenly all cellular networks went down and all GPS went down too the cars would choose a safe spot and pull over.
Quote:
Originally Posted by onewhippedpuppy
(Post 12510974)
How do you program a computer for every possible variable when sometimes things happen that have never happened before? Particularly for something as complex as a passenger airliner?
|
In AI we call these scenarios "the long tail" i.e. there are situations which happen to you all the time (approaching an intersection with a red light) and then situations which are rare to the extreme (airplane emergency lands in your lane on the highway).
One of the cool things about the current state of autonomous vehicles is that everything that happens with them (i.e. every bit of data coming from the hardware sensors) is recorded, meaning you can essentially replay and/or simulate anything that has ever happened to one of your vehicles. So if one of your fleet of autonomous vehicles encounters an escaped elephant running down the sidewalk of a major city, that is going to be flagged for "long tail" situations. Or, if the outcome of a driving situation is not ideal (for example the vehicle had to make an evasive maneuver which exceeded the parameters of what most passengers would find comfortable like a hard braking) that too can be replayed/simulated. As a result, the AI models can use these situations to continuously improve their capabilities and decision making. Then, even though only one car in the fleet ever encountered an escaped elephant, every car in the fleet gets better at dealing with that situation in the future. In the same vein, "what if" situations which have never been encountered "in the wild" can be simulated and tested.
Now consider this: you and I as human drivers may drive 1,000,000 miles in our lifetimes. Hopefully we get better as we get more experienced (putting aside for a moment that we will degrade in our old age) but that experience is just our own from those 1,000,000 miles. But autonomous vehicles learn how to be better drivers based not only on the experience of one car but rather from the experience of
all of the cars. Two years ago the company I was working for was the leader in the USA for autonomous driving, and we announced a major milestone of 1 million miles of autonomous driving in a single month. Now in 2025, Waymo is the leader and they are going to drive over
100 Million fully autonomous miles in San Francisco alone. The exponential curve is going to continue to climb, and with it the number of "things that have never happened before" and the cars being able to deal smoothly with those situations.