Tuesday 17th September 2019

Keeping the human touch in developing autonomous cars

Published on May 22nd, 2018

Phil Morse, international manager, Ansible Motion, explains how OEMs can use next-generation, simulated driving experiments to inform machine-based learning and build better autonomous cars.

With the ever-increasing complexity of today’s cars, OEMs in automotive remain on the look-out for potential efficiency gains in their vehicle development programmes. Things aren’t getting any easier; connectivity and autonomy means that the design, development, and certification of new cars won’t be any simpler or cheaper in the coming years. With more connectivity and advanced capability options filtering into our cars, how we react to technology interventions is going to become a crucial element for the acceptance of tomorrow’s cars. In complex traffic situations, sometimes the answer to how a driver or occupant reacts is not what one might expect. How might a real person, who is tired or distracted, respond to feeds from sensor-derived information and/or interventions in a vehicle’s trajectory? Can a connected car really deliver a pleasant, seamless and safe experience on our roads?

In the past we had to assess pre-production systems with real drivers in real cars. This was costly and required prototype cars, sequestered test facilities and proving grounds. These legacy options remain, but there is also a new alternative – driver-in-the-Loop (DIL) simulator laboratories.

Simulating reality

Even at first glance, these simulators offer vehicle manufacturers some obvious benefits over using real cars. They need to build fewer development cars, (which can cost $1 million each) and can test multiple scenarios in a safe, laboratory-like environment. DIL simulation gives ordinary people, not just professional drivers, the chance to experience potential scenarios that can be repeated in a controlled environment: there’s no need to worry if the weather has changed at the test track when it all happens indoors.

Further, DIL simulation has progressed dramatically in recent times. Legacy, ‘hexapod’-style simulators, derived from 50-year-old aircraft technologies, crept into the automotive space and have were the mainstay for the last few decades. Yet, the usefulness of these simulators has been limited because, it turns out, ground vehicles behave differently from flying machines.

So automotive technologies have migrated over time and now, in many cases, legacy aircraft technologies are being superseded by engineering-class automotive DIL simulators. They can be connected to hardware, even power-trains (the mechanism that transmits the drive from the engine of a vehicle to its axle). The simulators are so immersive that drivers respond just as they would in a real car. Hence engineers can recreate realistic scenarios to assess how people are likely to react to proposed systems.

Saving time

With connected cars, shifting more testing to the virtual world can reduce some of the inherent risk. Just consider Advanced Driver Assistance Systems (ADAS) and semi-autonomous systems, typically designed to intervene when the car is on the verge of losing control or about to collide with something. A US OEM has claimed that by pre-qualifying testing on a simulator, a ten-day testing session for electronic stability control (ESC) can potentially be reduced to just three days. That’s significant for any vehicle programme.

Reacting to the future

Alongside the more established fields of chassis engineering and human factors, it’s the new areas of electrification, connectivity and autonomy where feedback from human drivers is leading to more OEMs relying on DIL simulators. Test drivers’ behaviour can provide valuable and early insight into how actual drivers will react to the novel experience of intervention of autonomous technologies.

Such scenarios could be confusing or even dangerous if drivers don’t understand what the car is doing, and engineers can’t predict how they will react. Through early and regular contact with imagined systems in real-world environments, OEMs can use simulated experiments to inform machine-based learning.

Even the marketing teams can benefit; understanding how potential consumers interact with new infotainment features, data overload and drivers being distracted may improve customer satisfaction. And there’s also the actual human perception of the autonomous drive experience and how that fits with a car maker’s brand identity. Actual consumers, in meaningful sample sizes, can ‘experience’ real world scenarios and marketing teams can study them.

Looking further ahead, DIL simulation could answer other questions around autonomy. A simulator could be used to determine what causes, or to minimise, motion sickness, which autonomous car advocates are yet to understand. Ansible Motion is working with a UK university to investigate the phenomenon.

In the future, the way our cars work and how we control them is more likely to have been developed and refined in a virtual, rather than the real world. And few would argue that they won’t be better engineered because of it.

The author of this blog is Phil Morse, international manager, Ansible Motion

Comment on this article below or via Twitter: @IoTNow OR @jcIoTnow