The World Artificial Intelligence Conference 2020 is being held right now in China and brings together the best minds in the rapidly expanding industry.
With market leading autonomy, which relies increasingly on AI, Elon Musk provided a video during the opening ceremony.
During the video, Musk responded to some questions, the most interesting of which surrounds level 5 autonomy.
The question was:
How confident are you that level five autonomy will eventually be with us? And when do you think we will reach full level five autonomy?
To understand what’s being asked here, we first need to refresh on what is meant by each of the autonomous levels are, as defined by standards body SAE.
SAE Level | Name | Examples |
---|---|---|
0 | No Autonomy | Driver is 100% responsible for operating the vehicle. |
1 | Driver Assistance | Basic assistance – i.e. cruise control, lane departure warning, stability control |
2 | Partial Automation | Adaptive cruise control, parking assist, emergency stop, lane keeping assist, driver is definitely still required and responsible |
3 | Conditional Automation | Lane keeping, adaptive cruise, autopark, some level of hands-off/eye-off driving |
4 | High Automation | The car can navigate its way through most locations, but driver still needs to be present for edge cases. |
5 | Full automation | The car is fully capable of navigating and at this point steering wheel and pedals are not necessary. |
Before we get to Elon’s response, today Tesla vehicles sit somewhere between level 2 and level 3 when equipped with the Full Self Driving package that adds extra functionality to Autopilot.
This includes the car being able to change lanes by itself on the highway with Navigate on Autopilot, also the beginnings of low-speed transit with Smart Summon and autopark. In the most recent updates, Tesla’s now respond to traffic lights and stop signs.
This means the remaining challenge for Tesla is to piece together the high-speed (freeway), medium speed (city streets) and slow-speed (garage and carparks) all together to create one cohesive experience that enables the car to drive itself.
While that seems fairly simple as a goal, technically it’s a monumental challenge to deal with all the edge cases that are present on our roads and accommodate for the vast set of differences around the world.
Tesla realised what many other automakers still haven’t, that you’ll never code your way through that problem, you have to give the car a brain. That brain comes in the form of a connected computer that constantly learns and improves until it’s not as good as humans, but better.
So now for Musk’s answer to the question about level 5 and importantly, the timeline.
Ok, there’s a lot to unpack here.
When Musk says he feels like Tesla are very close to level 5 automony, there could be a couple of reasons for this.
Firstly, Musk is traditionally a fairly optimistic guy and we’ve heard some of these ambitious statements around timelines before. For example, at Tesla’s Autonomy Day last April, he expected it by the end of 2019.
Despite recent progress, I think the growing feeling among the Tesla community was that the months were ticking away and the revised end of 2020 goal was in doubt.
This new video really sets up a very interesting trajectory between now and the end of the year. Tesla typically delivers software updates around once a month, but judging from a recent Tweet, Musk expects a lot of functionality to arrive at once, somewhere between September and November.
There’s no doubt the Autonomy team, headed by Andrej Karpathy at Tesla are doing amazing things, albeit a fairly small team. Since Autonomy Day, the company has embarked on almost a complete rewrite of the underlying software.
This re-write is understood to include significant portions of the platform now being handled by AI. This means the data captured from millions of kilometres of human drivers, is condensed into a behavioural pattern that is expressed when your car approaches a similar intersection or segment of road.
Imagine you’re on a road you’ve never driven before, approaching a corner for the first time, you’re able to successfully navigate it at speed because you’ve got the experience of navigating similar corners in the past. Well thanks to AI, the car can as well.
As AI eats up more of the software stack (vision, drivable space, path planning, object detection, collision and obstacle avoidance, speed and sign recognition), Musk says the functionality of FSD will accelerate rapidly.
This acceleration could be what also gives Musk confidence, paired with the fact he gets to experience the latest development builds in his own car, well before even early access users.
If I’d driven in a car (or had the car drive me), it would change my perception of how quickly I’d believe this would be available, but I hope Musk is accommodating for that in his latest predictions. Of course, getting a car from one person’s home to their work without intervention is one thing, but to push out a software update and be confident hundreds of thousands could, anywhere in the world is quite another.
I hope the new timeline is to be believed, and when he says ‘I remain confident that we will have the basic functionality for level five autonomy complete this year’ that actually corresponds to level 4 on the SAE levels.
Listening to Musk talk about dealing with the long tail of odd-situations, I get the impression we’ll be at level 4 for quite a while, unless they can come up with an acceptable response when the car finds itself in one of these edge cases.
Imagine the car gets routed into a dead end. An example of this may be a series of roads that it had planned to take were then closed withing a short period of time due to a natural disaster like floods. With your car not able to navigate you out, it could have a failsafe of pulling over safely and calling Tesla emergency help. As long as these kinds of incidents were infrequent enough, it could be acceptable to the public, a bit like having a flat tyre.
The reason all of this is so significant is the massive opportunity that lays waiting for us when we can transport people and freight without costing lives.
The economics of shipping goods will substantially change and if Tesla are successful in being first, with great battery technology for range and lifespan, it really could transform how we think about car ownership.
If paying for a transport service (Uber without the driver), worked out to be much less than public transport or vehicle ownership would cost, then many, many less people would own cars.
Here’s the full clip of Elon’s speech to attendees, which runs for 17 minutes, or the whole uncut video from WAIC 2020 runs for 2 hours a 25 minutes. You can watch that here, but it is not in English and as yet does not have translation available.
If you’re into AI, driverless vehicles, or Tesla, then I highly recommend you give this one a watch.