The ‘autonomous dream’ is a long drive away, according to Five’s CTO
Don’t expect to see fully autonomous vehicles on the road in this decade – as with many complex software problems, self-driving cars are more likely to develop in increments, John Redford, chief technology officer at UK start up Five has predicted.
Speaking at a Dublin Tech Summit Session last week that looked at how close the world is getting to the self-driving dream, Redford -former CTO of Acorn computers – said that the hype cycle around autonomous driving was over and firms are starting to re-adjust their expectations.
“Everybody has realised it is a hard problem – probably one of the most complex and as yet unsolved problems of our time. So companies have gone from ‘drive anywhere in any sort of vehicle’ as their target to something much more specific,” he told DTS delegates.
“There’s clearly a genuine business need right now for removing the driver from long distance lorry driving, and so, people are targeting those specific niche cases. Or in China there are people developing 24-hour fully autonomous refuse trucks at a much lower speed use case,” he added.
For its own part, Five has pivoted from a self-driving car company to becoming a cloud-based development and testing platform for the software used in self-driving cars.
Redford explained the move: “When we started the company in 2015, we got to the point within three years of actually being able to do quite a good demo of a self driving car in London. We could drive 20kms in very complex London traffic with roundabouts, pedestrians – so we could make that work as a prototype.
“But we realised that the real problem is convincing yourself that you really could do this safely so that’s when we decided to start building tools to allow others to convince themselves of the safety of their system.”
Five’s CTO added that the other reason for the switch – and the reason the start up is in the process of being acquired by German engineering firm Bosch – is because self driving cars are a “multi billion dollar problem”.
He said: “You look at the big guys, they’re spending, maybe half a billion to $2bn every year on these programmes. To actually bring this to fruition, it needs a huge amount of money and it takes a lot of time. It’s big guys that are going to do that.”
While Redford acknowledged that specialist semiconductors are required for self-driving vehicles because of the machine learning that’s involved (Nvidia, for example, is a key supplier in this area) he doesn’t believe that the current worldwide chip shortage is a limiting factor in terms of where we are now in the development of self driving cars.
Nor does he think manufacturers are working in a particularly unfriendly regulatory environment.
“The UK is a generally a very friendly environment. And they’re also making the steps necessary to put the legal process in place as to who will be liable etc. And I think the same is true in a lot of countries because everywhere can see that it’s going to be a big business, maybe not right now, but 10 years, 20 years, this is going to be huge,” he said.
The other part of the puzzle that Five is working on is the vast amounts of data required to train autonomous vehicle systems and make them safer. According to Redford. Self-driving cars can collect 100,000 miles of data every week from their cameras and LIDAR radar data and they’re storing it and using it to train their systems.
According to Redford, Five’s platform analyses data from a fleet of test vehicles, using that data to create advanced testing scenarios.
The company’s engineers then build a simulation environment that makes it possible to assess and validate system behaviour at scale.
“Again, that’s sort of a more of a kind of a software technology company thing rather than a traditional auto thing. So I think these two groups are coming together in several collaborations,” he explained.
And like software, he thinks that developments in autonomous driving will happen in iterations. “You’re almost inevitably forced into a more incremental approach, where you try a set of things you see however it works, and then you learn and you fix the problem,” he said.
Consequently, he predicts that self-driving cars will experience a continuous if relatively slow rollout rather than happening with a sudden “big bang”.
“This decade for sure we will see real world applications but in terms of people driving around in driverless vehicles you are looking at the next decade,” he said.
Subscribe to our Editor's weekly newsletter