Computers are getting clever. But who is quicker, man or machine?
Is this pressure? I’ll admit to feeling apprehensive. I grab hold of the harness and pull my feet up the footwell, the light on the dashtop turns blue, the car emits an electronic fart, the steering wheel twitches and off it whirrs. Apprehension goes once we’re around Copse. Devbot 2.0 is a tugger.
I hadn’t really been worried about letting the car drive me around. Silverstone has plenty of run off and no one these days is going to put you in an autonomous racing car without being pretty damn confident in their technology. And their legal team. It feels way riskier taking your hands off the wheel in a Tesla.
Words: Ollie Marriage // Photography: Rowan Horncastle
But what if the car was faster? That I’d find harder to handle. But why shouldn’t it be? Supercomputers have been whupping chess grand masters for decades, and motor racing is a binary activity, with very defined track limits and a mathematically calculable fastest way around a circuit. Just depends how much you push the boundaries.
I pushed them as far as they’d go. Honour was at stake, there was no way the computer was beating me. This was a one lap challenge – man first, then machine. So for my single timed lap, I behaved in the most human way possible: went at it like a battle-curdled berserker. The full gung-ho lap – an if in doubt flat out, sort it out at the apex, route one, idiot’s guide to over-driving. Stick that in your algorithms and process it, machine. That it worked says much about both man and machine.
Let me tell you about Devbot 2.0. It’s the development side of Roboracer, the autonomous racing team whose Robocar has been performing around the world for the last couple of years. Performing, note: top speed runs, burn outs, donuts, but no actual racing. Yet. There are plans. They’re unconventional.
Devbot 2.0 (this is the second-generation model) is the serious side of the business: “Don’t think of it as a car so much as a hardware platform,” head of partnerships Paul Andrews tells me, “Devbot contains all the sensors, processors, Lidar, cameras and GPS which allows others to test their software packages”. Think of it this way. Anyone with an interest in autonomous driving, which is to say just about everyone these days, can use Devbot to test their software. VW used it recently to test a braking sensor algorithm. Once the VW engineer had uploaded the software package into Devbot, the Roborace guys asked him if he trusted his software enough to sit in the car while it did it. He did. It didn’t crash.
For car companies, Devbot can be a shortcut through development; for others, a blank piece of paper to sketch their ideas on. I meet Danilo Caporale, a postdoc researcher from the University of Pisa: “Our lab is a robotics and bio engineering lab. So we do a lot of things from robotic prosthetics to humanoid robots to industrial robots. They’re very, very different point of views. So we ask what can we do here that will improve the cars of the future? And by doing this, what can we learn to improve the robots of the future? Let me give you an example: we have a humanoid robot designed for disaster recovery in challenging environments, for instance a collapsed house. With Devbot we can test its sensors, how fast they respond, how they react in what we say is a very clean environment. So it’s not just about cars.”
Devbot hosts competitions for research companies, “not just track laps,” Andrews tells me, “but making the vehicle work out where it is if the GPS is disengaged or finding its own way through a maze of cones. The possibilities are huge”. Today’s exercise is comparatively basic, and so is the software it’s running – it can’t learn and improve lap by lap, just repeat. The race engineers tell me it’s unusual if the laps are more than 0.1 seconds different.
Strip everything away until you get to the chassis and you’ll find a Ginetta LMP3, but built up with an electric motor at each end, for 550bhp and 0–60mph in about three seconds. Here’s something I haven’t confessed already: we – Devbot and I – were limited to a hardly heroic top speed of 62mph. Not exactly full-blooded, balls-out racing, is it? But who does it actually favour? If I know anything from racing it’s that the faster I go, the more mistakes I make. Instead I start thinking tactics. Who needs the racing line? Faster, surely, to find the shortest route around the track. My brain goes into overload – these are parameters I’ve not thought about on a track before.
Distilled, it reads thus: flat-out, ignore the brake pedal, turn it in then sort it out. Same as always, but no need to run it out to the kerbs on exit. The steering’s heavy, so are the brakes, it pitches and jiggles and acts like the underdeveloped racing car it is. It’s noisy, too. Electric racing cars are, and not in a good way.
It’s the car’s turn. Less violent off the line, clearly attempting to feed the power in properly, not stomp. I know I’m a nose ahead already. At Copse it gives the inside kerb a foot of clearance, where I claimed a greedy three feet of paint. It’s good into Becketts, though. Brakes later than I did, applies them smoothly. Just doesn’t come off them nearly early enough. Or wallop the power back in properly.
It’s funny, though. Down Club Straight it makes the same jerky corrections it did on the approach to Maggotts, a binary jink and release like we do when we’re on Forza. In the corners it’s great though, adding steering progressively, balancing that with brake force. It’s very even-handed. But I know I’ve won before the engineers talk me through the telemetry and four-second time difference. Interesting looking at the screen traces, though – my apex speed was much higher, but the smoothness of the deceleration, the consistency of the steering, the slick inputs Devbot gave the chassis all suggest it could have gone a lot faster.
Later they try. An engineer is sent out with Devbot running a different software package, with a higher top speed and a more open attitude to risk. At Becketts it runs slightly wide, hits the wet white line on the outside and oversteers. Devbot catches the slide, but fails to hold it when it snaps back the other way. Yep, the robot car spins. This makes me happy – it shows fallibility, that machines have to learn too, and that that process isn’t an easy or fast one. But it, unlike me, will only get faster. Not sure I’m going to want to revisit this challenge in a decade’s time, but I’m sure as hell enjoying this moment while it lasts. Smug superiority, a very human trait.