Kyle Vogt, the CEO of General Motors' robot-taxi unit Cruise, has resigned from the company a day after apologizing to staff as the company undergoes a safety review of its U.S. fleet.
These self-driving car projects were always boondoggles. It always made more sense to work on standardizing road design and pedestrian handling versus trying to build some magic machine learning model that can perfectly manage the total chaos of humans on and around roads. Its like trying to build a program that can predict the outcome of a coinflip every tenth of a second for an hour.
These failed efforts have really done nothing besides cause regulators to throw up roadblocks and tank public perception for future self-driving tech.
Trains aren’t cars and cars aren’t trains. They can overlap but they don’t do the same thing. So if we assume we live in a world where cars and personal transportation will continue to exist (it will), innovation of cars will continue to make sense
I’m sure they’ve made strides and pushed tech forward, but until a general AI exists that can interpret novel/ incomplete info and make snap decisions correctly and consistently, this tech will never be safe enough.
What I find interesting is let’s say everyone had self driving cars in its current form tomorrow. Culturally, the underlying implication is that people are ok with giving up their autonomy with respect to driving their own vehicles. A pure machine learning based system in that scenario wouldnt even be ideal - it seems like it would be better to have a transponder type system where every car transmits their location to each other and to nearby control systems that also help direct traffic and car movement. Of course this means building up more infrastructure around traffic lights, stop signs, highways, wide open stretches of country roads, etc. but it feels like a more solvable and reliable solution than a ML based system, not to mention the fact that if we had 10 companies building self driving cars, every system would have slightly different behaviors
More than safety, the allure of self driving is that computers could smooth out traffic and speed up starting etc. And that is if the cars themselves can talk to each other.
If all cars could talk to each other they would be able to accelerate from a red light simultaneously (like a train), greatly decreasing transit times, as well as increasing throughput.
Same with merging and lane changes, they can be done precisely and smoothly if cars could cooperate. (Zipper merges would be magic)
Also, because the cars could theoretically know where everyone else is going, the can distribute traffic across multiple routes to optimize overall times.
And even in an instance where a car, say loses control, all the nearby cars could more easily avoid it without hitting a second vehicle. (and can not just let the car behind it know if it’s braking etc, but all the cars behind it, or even pass on relevant information to cars way behind, like if you get into bumper to bumper traffic, you can redirect vehicles that haven’t yet reached the traffic to reroute before you get there)
Even parked cars could all start to self adjust to increase parking, or to move out of the way of impact etc
These self-driving car projects were always boondoggles. It always made more sense to work on standardizing road design and pedestrian handling versus trying to build some magic machine learning model that can perfectly manage the total chaos of humans on and around roads. Its like trying to build a program that can predict the outcome of a coinflip every tenth of a second for an hour.
These failed efforts have really done nothing besides cause regulators to throw up roadblocks and tank public perception for future self-driving tech.
This problem was solved hundreds of years ago: Trains
But, that’s boring, though. People throw money at the new shiny stuff.
Trains aren’t cars and cars aren’t trains. They can overlap but they don’t do the same thing. So if we assume we live in a world where cars and personal transportation will continue to exist (it will), innovation of cars will continue to make sense
I’m all for innovation. That’s loosely what I do for a loving. I’m not for rushing to market.
Trains and light rail are mature products that can also be iterated upon with innovation. Imagine that.
I’m sure they’ve made strides and pushed tech forward, but until a general AI exists that can interpret novel/ incomplete info and make snap decisions correctly and consistently, this tech will never be safe enough.
What I find interesting is let’s say everyone had self driving cars in its current form tomorrow. Culturally, the underlying implication is that people are ok with giving up their autonomy with respect to driving their own vehicles. A pure machine learning based system in that scenario wouldnt even be ideal - it seems like it would be better to have a transponder type system where every car transmits their location to each other and to nearby control systems that also help direct traffic and car movement. Of course this means building up more infrastructure around traffic lights, stop signs, highways, wide open stretches of country roads, etc. but it feels like a more solvable and reliable solution than a ML based system, not to mention the fact that if we had 10 companies building self driving cars, every system would have slightly different behaviors
More than safety, the allure of self driving is that computers could smooth out traffic and speed up starting etc. And that is if the cars themselves can talk to each other.
If all cars could talk to each other they would be able to accelerate from a red light simultaneously (like a train), greatly decreasing transit times, as well as increasing throughput.
Same with merging and lane changes, they can be done precisely and smoothly if cars could cooperate. (Zipper merges would be magic)
Also, because the cars could theoretically know where everyone else is going, the can distribute traffic across multiple routes to optimize overall times.
And even in an instance where a car, say loses control, all the nearby cars could more easily avoid it without hitting a second vehicle. (and can not just let the car behind it know if it’s braking etc, but all the cars behind it, or even pass on relevant information to cars way behind, like if you get into bumper to bumper traffic, you can redirect vehicles that haven’t yet reached the traffic to reroute before you get there)
Even parked cars could all start to self adjust to increase parking, or to move out of the way of impact etc