Autonomous Nissan Leaf: riding in Britain's self-driving trailblazer

Details such as broken lane markings, cambers and road signs were all replicated in the model. An autonomous Leaf then drove through the real environment, with its trajectory and behaviour overlaid onto this digital visualisation. Data from human drivers following the same route enabled engineers to compare the cars’ trajectories, with the aim of fine-tuning the Leaf’s autonomous responses to provide a more human-like experience.

During our ride, the Leaf behaved naturally, driving around a parked car smoothly, slowing for corners, holding its direction through changing cambers that might have thrown it off-line, and accelerating and slowing progressively through changing speed limits.

At the end of the road, we approached a complex roundabout above the M1 with multiple traffic lights. With the first set of lights showing red, the Leaf pulled up smoothly before, once they’d changed, pulling away to the next set, also on red. Soon, we were clear to drive down the exit ramp to the motorway and the Leaf gained speed progressively before joining the traffic. Once on the motorway, the car held position but, unlike on the Grand Drive, avoided overtaking.

David Moss, Nissan’s senior vice-president of R&D, said developing a system that feels natural will be key to winning customer support for autonomous technology. He said the focus on improving the accuracy of the sensors means that “our system is much less reliant on road markings. We combined this technology with the way our own drivers performed to give the system more human-like behaviour.

“We won’t be able to map every road in anything like the same detail that we mapped the test route’s first three miles, which is why the machine-learning work at MUEAVI that enables a computer to recall and learn from scenarios will be vital to the system’s development.”

Professor James Brighton, a senior lecturer at Cranfield University’s Advanced Vehicle Engineering Centre, added: “We wanted to make overtaking as comfortable as possible, so we improved forward planning to make the process smooth and secure-feeling.”

Brighton said that although some autonomous cars overtake in the most efficient manner, it doesn’t feel natural, so “the autonomous Leaf positions itself earlier, follows a more natural overtaking line and leaves a comfortable space between it and the other vehicle”.

Autocar experienced this on the 0.62-mile-long MUEAVI test track, with a car parked at the side of the road. The Leaf began its manoeuvre in ample time and cleared the car with plenty of space to spare. Engineers are still working out how it will make the same manoeuvre in the presence of an oncoming car.

Brighton explained that human perception has played a major part in refining Nissan’s and Hitachi’s AV systems. For example, a narrow width restriction makes drivers slow down automatically, whereas an AV won’t, potentially upsetting passengers. A high-sided vehicle that’s passing will make a human driver instinctively move aside but an AV will continue on regardless.

Brighton said those differences in behaviour mean that, to create an AV that drives like a human, “we had to investigate and consider drivers’ attitudes to a car’s longitudinal velocity and lateral position on the road”.

Moss said the Grand Drive trial was a great success and rejected claims voiced elsewhere in the AV industry that driving around on public roads teaching autonomous systems how to react to situations is dangerous, inefficient and ineffective.

“Safety is our primary concern,” he said. “Over 93% of accidents are caused by human error. The technology we’re developing will make future cars safer while elements of it will soon find their way into our production cars.”