It’s about the integration, stupid
Jaguar Land Rover’s senior manager of autonomous driving systems, Isabella Panella, believes the self-driving sector can learn from the disciplines of UAV production in the aerospace and defence industry. Interview by Ray Molony, technology editor of The Autonomist.
As the UK’s largest automotive manufacturer and the owner of two of Britain’s most iconic car brands, Jaguar Land Rover is the flag carrier for the country’s advance into autonomous vehicles.
Its pioneering deal in March 2018 to supply 20,000 of its I-Pace electric SUVs to Waymo’s fleet raised eyebrows – and JLR’s international reputation – as did the completion of the first ever self-driving lap of the challenging Coventry Ring Road by a Range Rover Sport in October.
The company spends around £4.5 billion a year on new product creation and capital expenditure and has famously announced that electric versions of all its vehicles will be available from 2020.
An exciting time then, to join the company for aerospace engineer Isabella Panella, who became JLRs senior manager of autonomous driving systems in November.
A self-driving Range Rover Sport completed a real-world test on complex Coventry Ring Road last October, successfully changing lanes, merging with traffic and exiting junctions at the speed limit of 40mph. Pic: JLR
Apart from a brief spell at Rolls Royce, Panella’s background is in aerospace and defence, specialising in unmanned systems, including military drones.
Panella, an Italian who graduated from RMIT in Melbourne, believes the embryonic autonomous sector can learn from some of the more established disciplines of software integration in the aerospace industry.
‘In the automotive sector, there’s an underestimation of the importance of system engineering.
‘System engineering isn’t just a word, it’s a way of guaranteeing traceability and robustness.
‘When you design an autonomous system, especially with the variability of the behaviours you can expect to experience, you need to be able to justify the ‘why’ at every point in time.
‘Why did it behave like that? Why did you design it like this? Otherwise, you’re liable to be jailed, in simple terms.
‘As part of this process you need to have traceability of all the requirements and prove that you’ve done your due diligence at the engineering level.
Panella believes that we need to migrate more into a virtual tasting environment, where we inject fault in the system – variation and deviation from normality – in a very short timespan and see the reaction of the intelligent controller.
‘So in my mind that’s one of the reasons that JLR brought me in. My knowledge of software, systems engineering and autonomous system integration. It’s not just about designing them but implementing them in the real world.
‘It’s something that the automotive industry can miss, as it’s engineering focused rather than software focused.
‘The complexity of autonomous systems is based on the interdependency and interconnectivity between different technologies and some of those technologies need to be carefully designed to make sure that we have an integrated system that can work within a commercial environment.
‘I think there’s usually a tendency to have ad-hoc solutions rather than an integration solution and that’s why we end up wasting a lot of time integrating them and trying to make them work. In order to be successful, different industries need to come together.
‘In any autonomous system there are three key pillars: one is safety, because that drives the requirements at a high level; we have sensing, to create the perception of the external world in order to support collision avoidance or reaction and proactive behaviour in that external world; and then there is autonomy in the form of machine learning or some sort of processes that can enable us to flexibly adapt to the different circumstances that an autonomous system would experience’.
A surveillance UAV, the General Atomics MQ-9 Reaper. Panella believes the self-driving industry can learn from the disciplines of UAV development and production. Pic: US Air Force
Panella is concerned that the players who trumpet mileage achievements such as the 10 million or 1 billion miles of self-driving are missing the point.
‘Even if you drive forever, there will be circumstances that you cannot have foreseen. In my mind, it’s not about how many miles you drive, it’s about how you create reliability and robustness in the system and I don’t think we’re focusing enough on what characterises reliability in autonomous systems.
‘We should be focused on collision avoidance, percentage of collision and probability of collision because that’s what it’s about’.
What really concerns her however, as a relative newbie to automotive, is the lack of clarity over autonomous standards.
'We still don’t have clear parameters for safety and regulation in the automotive industry and in my mind that’s a very dangerous game because we cannot assume that what we do and what we invest in is going to be robust enough for future years.
‘You need to have engagement with the safety and regulatory bodies and make sure they advance their regulations as soon as possible. We can’t just be sending robots out there…’
She believes the verification and validation process could be more demanding.
‘I think we need to migrate more into a virtual tasting environment, where we inject fault in the system – variation and deviation from normality – in a very short timespan and see the reaction of the intelligent controller’