Slashdot Mirror
Google Has Stopped Developing Its Own Self-Driving Car - Report (techcrunch.com)
Google has reportedly shelved its long-standing plan to develop its own autonomous vehicle in favor of pursuing partnerships with existing car makers. From an article on TechCrunch: The Information reports that Google's self-driving car unit -- known internally as Chauffeur -- is working with established automotive names to develop cars which will include some self-driving features, but won't ditch the steering wheel and pedal controls. The firm is already working with Fiat Chrysler, per a partnership announced in May, and that could be the start of others to come. Google first set out to do away with the steering wheel and pedals approach, but this backtrack is from Alphabet CEO Larry Page and CFO Ruth Porat who found the original approach to be "impractical," according to the report. That's despite Google's autonomous vehicles clocking over two million miles of tests on public roads.
How many self-driving cars went into tunnels and stopped when they lost GPS and Internet connectivity?
None. They cache the GPS + map data for several miles in all directions for this type of scenario.
For example, if I'm going to an event in a rural area I'm probably going to have to park in an improvised parking area on an unimproved or only marginally improved surface. I may have to drive down a trail that itself is unimproved or only marginally improved, either following the directions of humans waving at me or else following something like the occasional orange cone or even the tracks of previous vehicles.
Heck, this is pretty common in an urban environment. If there is utility work going on, you'll see a few cones strewn about to vaguely indicate you are to use one of the oncoming lanes, with a cop looking down at his phone waving at you desultorily.
Steering hasn't really worked that way for decades. Under the dasboard, the column bolts to structural elements. A shaft from the column enters a universal-joint and then continues through another shaft to another universal-joint, then through the firewall and down to the rack and pinion bolted to the body of the car, which in-turn moves the tie rods to push and pull on the steering knuckles. Automakers usually introduce a series of rubber isolators through the shafts to further reduce vibration.
This system helps to protect the driver from being impaled on the steering column in a severe accident, the u-joints will allow the deforming car body to not press the whole column right at the driver, the part bolted to the dash remains bolted to the dash while the shafts bend around each other in a sufficiently serious accident.
You're probably going to get as much feel through the vibrations from the suspension through the chassis into the body then through the dash to where the column is mounted as you would through the shaft.
Do not look into laser with remaining eye.
I've read various estimates that it takes a human somewhere between 5 and 17 seconds to take over from a self-driving car when notfied, when they were concentrating on something else.
So this poses an interesting design dilemma. If you put in a steering wheel and manual brake pedal etc, and have a situation requiring emergency rapid action, and the automation system is in the middle of taking the action it computed is best, how do you PREVENT the human from providing contrary control input which in all likelihood will mess up the overall response to the situation, especially since they are very likely coming in way late.
In what circumstances do you keep the human input disabled, for reasons like mentioned above, and in what circumstances or after what delay do you let them take over. A combined control-input situation would be disastrous, like having the "backseat" driver sitting beside you grab the wheel in panic while you're in evasive driving.
Where are we going and why are we in a handbasket?