I get there are car AI systems that can take you from point A to point B under controlled open highway driving circumstances, but in various articles about Tesla and Uber self driving cars I’ve seen lately there seems to be the assumption this will be a pretty near term (within 5 years or so) scenario for city driving.
My Galaxy smartphone attached to googles giant mega network can’t even do voice recognition properly and we’re banking that a car will be able to drive itself safely in the complexity of city traffic? I think of tech like this as being 10 - 15 years out at best in terms of actually being safe and effective.
Possibly due to Tesla, we now have marketing momentum, where every company even remotely connected to the automotive industry has to announce a self-driving something. We’d have to read the fine print to see where it can go and where it can’t, not to mention when.
My guess is that some companies are going to release them, some people will drive them, and it will all pause when a car causes a 20 car pileup or something - something much worse than the guy who killed himself watching a movie, not the road. Then we’ll get some regulations.
However with all these resources thrown at the problem it will be faster than 15 years for sure.
different problem set. you’re falling into the “we can put a man on the moon” trap. Voice recognition is hard because computers can only work on the total sound they hear, which is your voice plus all of the other noise around it. if the noise is great enough to obscure the waveform generated by your voice, it’ll have a hard time. plus, from a language point of view, a computer can only do what you say. It can’t interpret what you mean.
it’s already about 70% there. If you load up a car with adaptive cruise control, auto park assist, lane keep assist, blind spot monitoring, and forward/cross traffic collision warning, then you’ve got 70% of a self driving car already. it can steer itself, brake itself, hold a lane (most of the time,) etc. The last 30% is going to be a bastard to get right, but it’s possible. advanced sensors like LIDAR are going to be key; current assistive technologies like I just mentioned can only identify that an obstacle is there, and usually how large the obstacle is. They can’t identify what the obstacle is. I saw a demo of a car equipped with a full suite of lidar sensors, and they had a screen showing a stitched together video of what the car’s LIDAR sensors were seeing. you could easily see what was a person, a parking block, and so on.
Five years is grossly optimistic for truly automatically piloted cars, but not only is the technology self-piloting cars viable with reasonable advances in machine intelligence and vision, it will likely become the default mode of automotive transportation within a generation, and being allowed to manually operate a car on roads will become the exception if permitted at all.
The problem with implementing self-piloting automobiles right now is twofold; one is that will the state of the art in self-piloting cars can cope with normal driving conditions and situations, the remaining small fraction of a percent of situations where it fails occur often enough that automated vehicles represent a significant hazard. Of course, human drivers also represent a significant hazard as well as they are frequently distracted, impaired, or just plain incompetent at driving. However, human drivers have personal liability defined in law, whereas it is unclear just who would be responsible for property damage, injury, death resulting from a self-piloting car, especially in a scenario in which the automaker or operator could not reasonably be held at fault, or an owner modifies or fails to maintain a vehicle. Establishing basic standards for how automated vehicles should operate and what kind of safety principles and restrictions should apply will be a learning process for both industry and government.
The second problem is more of a transitional one; as automated vehicles become more capable it will be the irregularity of human drivers that represent the bigger hazard. It may actually make sense to change basic ways in which a vehicle operates; for instance, instead of having stoplights for cars going opposite directions, it is much more efficient and safer to have the incoming vehicles communicate with one another, or with a local “monitor” to negotiate speeds such that they pass through the intersection missing each other but not having to stop or slow down excessively (which is one of the biggest causes of reduced fuel economy and efficiency). However, a human driver cannot be relied upon in this system, so the system will either have to make special accommodations to cope with the remedial human driver, or else human-piloted vehicles will have to be capable of an override in such cases to allow it to be guided through the intersection. Human drivers will naturally resist this kind of interference, but given the large number of accidents that occur at intersections it is most sensible to automate this kind of interaction. One of the problem with “driver assist” features like Tesla’s poorly named “Autopilot” is that it allows the driver to not pay attention but can actually increase driver workload by needing to pay attention to another system, making driving less safe. But then, many drivers are already operating vehicles as if they are self-driving by focusing attention on media or texting.
This transition won’t happen in five years (and the technology won’t be there to support it in that period) but I’ll be surprised if the self-piloting car isn’t the default within twenty. There will still be certain applications where manual operation of certain classes of vehicle will be necessary but it will be the exception, to be performed by only specially trained and certified individuals will be permitted to operate vehicles on public roads.
As I’ve posted elsewhere on the Dope: Autos are already on the road in Singapore; they are taxis developed by a company called nuTonomy. They began service Thursday; the future is now.
This commercial amuses me. It’s from a law firm soliciting for people to join its class action suit against a pharmaceutical company. The pitch? “Women who have experienced permanent hair loss after chemotherapy may qualify for compensation.”
I thought it was common knowledge that one of the side effects of chemotherapy was hair loss?
wrong thread … but there getting sued for the fact that they didn’t tell every one it might become permanent which is normally rare but seems to be more occurring with that particular company
Unfortunately, one of the normal driving conditions, at least for the upper half of the country, is snowy weather, and they’re not currently capable of handling that.
Well, yes and no. Computer controlled traction control is far more capable of maintaining traction and recovering from loss of traction conditions. Where an autonomous vehicle will have problems in recognizing road markers on a snow covered road, which is something a driver could assist with, and heuristic systems can learn to recognize signs of roads from the response of a human driver. It’s a challenging problem, but capable of solution. What will be more challenging is predicting the often erratic behavior of other (human) drivers with whom it cannot communicate.
Where does GPS fit into all this? It’s not uncommon for GPS systems to direct people to the obviously wrong place. How would this be managed in a self-driving car? Would there be some kind of override?
I think the question of “when” is currently more about politics and consumer demand/perception than technology. The google car technology already has millions of miles under the belt and driverless car laws are being passed in multiple states.
Issues with confusing GPSs, horrible intersections (I’ve been through a few I don’t think have a deterministic solution) etc. can be solved rather easily, it’s just a question of getting the data in and having the incentive to implement the solutions.
I think the inflection point will be when they start being available in car pooling services. Those are currently limited by how many locations they can have for their cars. I personally wouldn’t sign up for one if I had to walk more than five minutes. But if the car would deliver itself to the door when I needed it?
As we say, the first 90% of a project takes 90% of the time. The other 10% takes the other 90%. I think we may be in that final 10% now.
5 years is too soon for them to be certified and fully available, but 15 years is too long.
I think with the current data collection capabilities that’s easily solvable. GPS will never be accurate enough, especially with buildings casting satellite shadows, but these cars have cameras and radars than could see the surroundings, even when visibility is zero for humans. Build up google maps street views with all the stuff you’ll still see when there’s snow and they could easily triangulate the exact location of the road.
I saw one of the first CNC machines, circa 1974.
It was a milling machine - programmed by having a human perform the task and recording the motions.
With the Cloud (All Hail the Mighty Cloud!) as a database, it may actually be possible to record the road as driven when clear and dry. And pay special attention to this intersection - it’s a killer!, thereby creating a massive database for the car to access when things go belly up due to weather.