There are an extraordinary number of companies converging on the idea of self-driving cars from all sorts of different angles, but the undisputed leader of the pack is Google, whose self-driving vehicles have logged a total of nearly 700,000 development miles. In a video released today, Google shows some fantastic visualizations to demonstrate what the car is seeing in complex traffic situations. You can watch how the car handles roadworks, level crossings, complex intersections and a range of interactions with cyclists – including recognizing their hand signals.
The images in the gallery show what Google’s Self Driving Car is seeing in a range of traffic conditions through its 360-degree cameras. Let’s take a look at some of them in closer detail:
In this image, the car has spotted a roadworks scenario blocking the road. It sees the signage in yellow, plus a bunch of cars in purple, and a group of witches hats in orange. It’s approaching the scene with caution and has not yet made a decision to proceed – the red line indicates a barrier the car will not cross.
In this image, the car is adjusting its lane position to give a wide margin to a parked car.
In this image, the car has recognized a rail crossing, and the red bar shows that it won’t begin to cross the train tracks until the traffic is clear on the other side.
This image shows the car recognizing a cyclist’s hand signal indicating that he’s coming out into the center of the road. The green bar shows that the car won’t pass the cyclist under these conditions, and the other red object is another cyclist traveling behind the car.
This image shows the car dealing with a typically complex city traffic situation. As the lights have gone green, a number of pedestrians and cyclists have all started to move through the intersection. The car sees and tracks each of them individually, stopping at the intersection until all cyclists and pedestrians have moved through, and even waiting for a cyclist behind the car to come through rather than impede its path.
With 360-degree vision and non-stop eagle-eyed attention, the Google car is already proving a far safer driver than most humans – at least, under the conditions it’s currently being tested under.
But don’t hold your breath – it’s still a long way from being a product you can buy, and there’s a range of issues such as localization, licensing and liability, regulatory approval and general public opinion to get past before self-driving cars become a commercial reality on our roads.
Still, this is some seriously amazing technology, and driverless cars are more or less an inevitability in the coming decades. Safe, cautious and courteous to a fault, they won’t be the fastest way to get from A to B, and they’ll probably be as annoying to be stuck behind as an overly cautious octogenarian - but they will certainly save lives. At least, until somebody develops a way to jailbreak the software and install some Ken Block drifting firmware.
Source: Google
Before self driving cars are ready to be let loose on the roads the individual components like this that make up the platform need to creep their way into human controlled automobiles.
I am far less convinced than most other people that the technology is ready to be let loose on the roads without direct human control or someone perpetually ready to override it.
Maybe its just because I see regular failures in much simpler technologies that I have very little faith that such a platform could be made mature in the near future. The Google cars have a very good driving record so far but engineers ride along with override controls and the company doesn't (or didn't) publicly disclose how often those controls are used.
I think the "hybrid" approach has far more potential. Simple actions like if a driver is still accelerating and the car in front of them stops and they appear not to notice, the car and switch from gas to gently braking and provide an audible alert to the driver to observe the situation.
Another useful situation in rural areas would be using Infrared to identify animals near the side of the road I may not be able to see on my own or giving me a birds eye view of my lane position and surrounding cars for checking blind spots.
Once these augmentations become established as reliable it would then make sense to remove the human and allow them to operate independently but not before then.
There are some other rare situations which I am not sure self driving cars could avoid entirely as well. I remember one time I was in my car and a kid was driving really fast on the sidewalk coming off a hill and I wasn't sure if he was going to make the turn. I backed off and moved over just in case, he hit a bump and flew over his handlebars into what would have been my path. I am not even sure where you would start in attempting to program the car to figure out "this kid looks like he might dump it".
I realize circumstances of human error are almost certainly more common than the one offs like that automation would judge poorly but at least for my personal record there are at least 2-3 times the machines would have done worse than me and just one where autopilot would have avoided an accident I didn't (I was driving way too fast on a slippery road I didn't know).
The problem with large container subway type transports is they must stop frequently and they are limited in reach. I think they need to be broken down into a much smaller "packet" type of system so there is no need for the whole system to stop for individual "small containers" that need to get on and off the system.
The small containers could reach smaller access roads instead of just large main transport lines so they could get closer to the destination than a subway could. This same principle eliminates the need to travel in a car to a subway station.
When you get to the destination you don't have to worry about finding a place to park the small container as you can just send it back through the system for someone else to use (sort of like a ski lift).
Because each of the cars has sensors and the grid itself has intelligence its easy to measure delivery times, reroute pods around congestion etc.
In theory because the pods are likely electric and the grid provides power its probably possible for them to travel short distances off the designated "track" on just battery power. If they meet standard spec/protocols etc. they could even be personally owned like vehicles.
Because of efficiencies and being able to travel much closer together while synchronized on the track and not needing to fully stop I think you could achieve much greater throughput than is possible with either public transport or conventional cars.
Now that we have detailed urban environments in sandbox type video games I would really like to see this concept built virtually inside of one.