(slightly nervously) If it did then that’s the moment when the machines have taken over, right?
Well, we have looked at a way basically of… it’s another challenge, with AI. The whole point of Roborace is that we’re developing AI technology and one of the things is how does the car, a future road car, communicate with the occupant? A lot of research is going on in terms of voice communication, so how does the car talk to the human to engage the human and inform them of things that might be going wrong technically. So rather than having a warning light appear on a dashboard, it actually just tells you.
Like Siri or Alexa?
Yeah, so it’s able to interpret data and communicate it. So what we’re looking at in that example of one car’s hit another is ok, so now the car itself has to interpret its own data and then give a justification for what happened.
So it might say ‘Robin’s driven into you’ and Robin’s car might be saying ‘Seb’s driven into you’
Exactly. So one of the interesting things that we can do is we can have an insight into the mind of the driver live during the race.
By the way it’s processing data?
Yeah, so what we say is we want to know what risks the vehicles are taking. So when a car makes an overtaking maneuver down the inside, what’s the percentage chance of contact or a successfully completed move? So you immediately know whether it’s like 80% chance or 50% chance that there’s gonna be contact. But what also happens is that the car in front is also making that assessment about being overtaken and about whether this car behind is actually gonna achieve what it’s trying to do or not.
And that’s the only way it would ever be able to avoid an accident, by staying wide, letting the car outbrake itself and then cutting back on the inside. All of that logic is to do with risk and that’s what racing is all about. How do I control your behaviour, how do you react to my behaviour, it’s all prediction. And when you’ve got two different AIs that are competiting, that’s the real challenge.
There’s no difference then to how do you merge on a roundabout, how do you merge on and off a motorway, it’s the same challenge. You’re trying to influence other objects’ behaviour. So purely doing what we’re doing at the minute and driving round on our own on a track – that’s the foundation, that’s the perception layer, that’ about the accuracy of your location within an environment.Once you have that, that’s your foundation – then you start to move into vehicle interaction.
So you can actually see them race each other. I know there are now two Robocars…
We actually have two built – three by the summer, the third one is being built at the moment. And then we have three Devbots as well.
Oh, I had no idea there were multiple Devbots
Yeah yeah, so we have three Devbots, which makes it six vehicles that we can test and develop in. Which is really important from a software perspective but also then for looking at interaction – you need multiple objects to do interaction, obviously.
In terms of the actual Robocar, obviously Devbot is a real car that human racers drive, whereas Robocar is not designed to ever have a human in it-
So there’s no driver cockpit in Robocar, no.
But it does have – from when I’ve seen it at ePrix – a sort of cockpit for the AI?
(Laughing) Yeah yeah, there’s an AI service hatch – so effectively the only difference is that the “eyes” of the driver are at the front of the car, they’re not as a helmet that needs to stick above the bodywork. So that’s pretty much the only difference.
Yeah because it doesn’t have the rear wing, because it has such a low profile?
Yeah, it doesn’t have the rear wing and we do most of the aerodynamics through ground effect, from the large diffuser at the back. There’s a small rear wing at the top of the diffuser, which adds to the downforce and there’s two front splitters as well, on the left and right of the car.
I read that it’s 975kg, which is nearly 100kg heavier than the minimum weight of a Formula E car, including driver. To a lot of people that might seem really wild but presumably there has to be a lot of pretty serious tech in there to physically move brakes and things, even if it’s run by a computer?
So in terms of weight, we’ve got a battery that’s twice the size of a Formula E car’s.
Oh, ok. That would do it then!
And we’ve got four motors where they’ve got one. So a lot of the weight comes from those components. The reason we’ve gone for four motors is we can do torque vectoring on the car. So we can do very advanced vehicle stability.
Sorry, just to check – there’s a power train on every wheel?
Yeah, one motor per wheel, fully independent. And when we’ve tested with that just by implementing torque vectoring we can save about two seconds per lap, at a Formula E track.
That is a lot.
That’s with no other setup – that’s just about power deployment, so you’ve not improved your motors, you’ve not improved anything on the power train, you’ve not changed the battery, all you’ve done is just deploy torque in a very different way. So it’s like an intelligent use of the energy that you have available. And it’s worth two seconds a lap.
It also means on road cars you’ll be able to control the car much better in extreme situations. So when they have things like the Elk Test, where an elk comes out and you have to swerve to avoid it or a tractor reverses out and you have to evade it, that type of dynamic movement is where torque vectoring can really help with vehicle stability.
So you said it’s got a battery twice the size of a Formula E car, especially allowing for the fact that with four power trains you can have a lot more regeneration – or will the Robocar manage regeneration as well? Because for the human drivers that’s very complex.
Yes, so, it’s really interesting – we have regen and we have regen on each motor. When you say that regen is complex for the driver, it’s kind of interesting because now actually there’s a computer that’s in control of that.
So they have a brake pedal and they can adjust biases but effectively they’re not responsible for doing the full torque distribution anymore. It’s mixed between the human and the computer system. So it’s another example of where technology is creeping in and taking away driver skill – if you actually gave a driver the ability to have two brake pedals, one for regen and one for manual braking so then you can balance those two things then you’re back to a purely human competition. The moment you say a computer does it then the human’s not doing it anymore.
So I find that really interesting. Again, you look at human vs. robot competition, there’s a lot of things even in modern racing where actually the computer’s doing the work for the human.
(Editorial note: I think the drivers would disagree with Bryn about this, from a very thorough lowdown Jerome D’Ambrosio gave me of what they’re controlling in the car. I am certainly not the person to arbitrate that pitlane quarrel, however!)
I saw di Grassi talking about – obviously, he’s very enthusiastic about ROBORACE – talking about the fact that particularly the Endurance cars have huge amounts of software in them and I think people don’t realise quite how much spec software goes into even the GT cars to turn them into race cars, let alone the Prototypes.
It’s really interesting because you come back to the human-machine interface with a vehicle, which is basically a steering wheel, a throttle and a brake. And that’s pretty much what we had and we had that in the 1800s. When the car was first designed – well not when it was first designed, there was lots of different variations then – but the conclusion was, a bit like having a mouse and a keyboard, it’s a steering wheel, throttle and a brake, they’re mechanical systems.
But with that you have a lot of limitations. In that sense, you can’t individually control brakes across all four corners, you can’t individually control torque distribution at all four corners. On braking you can’t change how much you use mechanical brakes compared to regeneration. All of that has moved into a software domain. But the really, really interesting thing in the long term is: is a steering wheel, throttle and a brake the best human-machine interface we need to control racing cars of the future? If you want to have a human vs robot competition, there has to be parity in the control that the human and the software have of the vehicle. If you just give the human a steering wheel, throttle and a brake you’ve already hampered their capabilities.
Because they’re gonna be slower at managing that, than the machine?
Yeah, effectively you have one torque input. So the example being if you put a human in Devbot without torque vectoring, just give them one throttle that you allow to be evenly distributed then he’ll be those two seconds a lap slower than an AI that is able to use torque vectoring.
If you enable torque vectoring for the human then, well, he’s not in control of it anymore – the software’s in control. So that two seconds a lap has come from giving the computer control, not the human. So it’s not a human vs robot competition.
No, it’s a collaboration.
Yeah – so it’s working out that purity of what do we want human racing to be in the future. And we’ve touched on it before in things like Formula 1 where we’ve had fiddle brakes on the cars. McLaren had a “secret” braking pedal or lever, which was designed to help with torque vectoring at the back of the car, effectively a torque vectoring type system.
But the FIA banned it. They banned it as a driver aide when it’s not actually a driver aide, it’s a control interface. It’s actually a driver skill – to be able to control the vehicle you had to do something and that’s really interesting, it was something that I don’t think should have been banned at the time that is exactly that example of if you have two brake pedals then it needs driver skill to use it.
Is the Robocar geared? I’m always kind of interested by this in Formula E because a lot of them cling to the concept of gears, which the electric motors don’t need.
We’ve got a single gear, so we have a step-down ratio and we use that. But you are balancing torque demand and final velocity, as well.
Thank you for talking to me for so long, by the way, this has been really great. Do you have a timeline at the moment of when Robocar will go into full-scale manufacture?
At the moment that’s not confirmed, lease has not been signed so can’t really talk any more about that. So as a manufacturer we use a lot of partners to manufacture the chassis – we actually assemble.It’s all designed by us and then manufactured by partners that we use. So we do final assembly of the vehicles and at the moment we do that in Yarnton, in our factory.
Are we gonna see the Robocars on track in the next year, do you think?
Obviously we’ve had Robocar on track in Paris this year. And then we’ve actually had the two cars in private testing – we use a private testing venue as a replica Formula E track that we’ve built and we use that for all of our software development and vehicle development and that’s where we’ve been for the last four or five months.
So it’s pretty close to actually racing?
The target is really for December, Hong Kong, for Formula E and working out what do we put on track then. We’re in discussions with Formula E about the best format, the best entertainment that we can provide at the venue but also having the ability to communicate what autonomous technology is; there’s an educational piece that we need to do. Rather than just going “oh, there’s a couple of cars on track,” it’s like how do we tell the story about autonomous vehicles? Formula E have done a great job with that for electric power trains, we’ve got to do the same for connected, autonomous vehicles.
So pretty much the same communication job you need to do between car and driver, you need to bring to the general public?
Yeah, exactly and that will take time. But it’s a really interesting challenge that the industry is going to face in like 2, 3 years time. Actually with Audi it’s coming really soon…