Driverless cars are on their way – there’s little doubt about that. But before they hit the UK’s roads, they need to be tested in realistic simulations to ensure that this transformative technology will be a safe and positive addition to our lives.
Latent Logic, an Oxford University spinout company, is helping enable this vital testing with technology that teaches autonomous systems using real-life examples of natural human behaviour on the roads.
Professor Shimon Whiteson of Oxford’s Department of Computer Science, co-founder and chief scientist at Latent Logic, explains: ‘Autonomous vehicles must be tested in simulation before they can be deployed on real roads. To make these simulations realistic, it’s not enough to simulate the road environment; we need to simulate the other road users too: the human drivers, cyclists and pedestrians with which an autonomous vehicle may interact.
‘Latent Logic is using a machine learning technique called imitation learning to build realistic human behaviour models. These models make it possible to test autonomous vehicles quickly and robustly.’
Latent Logic grew out of an EU-funded research project that trained semi-autonomous telepresence robots to behave in a socially ‘normal’ way. Since it’s difficult to quantify what is meant by socially normal, it’s much easier to train such systems to imitate the behaviour of humans acting in a socially normal way.
Professor Whiteson decided to explore the commercial potential of this technology, recruiting postdoctoral researcher Dr João Messias to be co-founder and chief technology officer. Kirsty Lloyd-Jukes then joined the company as CEO.
The technology works by combining state-of-the-art computer vision with imitation learning. Professor Whiteson says: ‘Our models extract the “latent logic” behind real-life examples of natural human behaviour. As a result, they can respond realistically even in new situations.
‘We use computer vision to collect these examples from video data provided by traffic cameras and drone footage. We can detect road users, track their motion, and infer their three-dimensional position in the real world. Then, we learn to generate realistic trajectories that imitate this real-life behaviour.’
By providing a service that enables better training and testing of autonomous vehicles, Latent Logic hopes it can hasten the safe introduction of what will be life-changing technology. Professor Whiteson adds: ‘While many of the biggest players in this market are international, there is also a lot of energy in the UK in this sector, and Oxford is a hotbed of talent and entrepreneurship in machine learning, robotics and autonomous vehicles.
‘Autonomous vehicles are improving rapidly, but we are still some way from realising the dream. It is not just about perfecting existing technology – there remain fundamental unsolved problems in building sufficiently robust autonomous systems. At Latent Logic, we believe our technology can play a critical role in addressing those unsolved problems.
‘In addition, our technology has numerous other applications. Situations in which you want socially normative behaviour are great candidates for learning from demonstration, as is robotics – from factories to warehouses to homes. You can also think about video games, where people might want to play against bots that can imitate the style of their favourite professional gamers. The sky’s the limit.’
