Cambridge UK technology great Arm and the company’s ‘godfather’ Stan Boland have called for less hype and more reality in the drive to get safe and reliable autonomous vehicles on our roads.
A damning report, commissioned by Arm, says key components must be “almost military-grade robust’ and adds that public expectation that driverless cars would be completely safe – “perhaps 10X as good as the average human driver – could be as far fetched as dealing with a meteor falling from the sky.”
Boland, who was brought in by Acorn Computers in 1990 to extricate the embryonic Arm from the quoted group and into separate ownership, commends the Arm-Forrester Consulting report and says major players in the AV market need to address a clear disconnect between propaganda and the safety issues at the heart of the debate.
Boland is now CEO at FiveAI, a Cambridge company bringing together the best minds in AI, engineering and mobility to deliver a fully autonomous shared transport service for Europe’s cities. It is managing rigorous off-road testing of trial vehicles.
Boland told Business Weekly: “No question that any serious firm aiming to bring self driving technology to market knows that this is all about achieving safety assurance.
“Systems we are all working on must deal with a virtually infinite possible set of, and combination of, objects, states and behaviours and do so without the general intelligence we humans routinely apply.
“That spans from tasks such as to detect all the salient objects and features of the scene, to classify them, to determine distances, orientation and states, to make sense of the scene, to predict human actions and reactions, to determine uncertainty and to plan a path and carefully control the vehicle on that path. And to do so with a reliability that is at least as good as a human – in the long-term much better of course.
“It starts with understanding and analysing the target domain – that’s a very geo-specific effort but an expensive one. Then each component (hardware or software) needs to meet a specified performance guarantee (functional safety).
“We need adequate redundancy in case of component failure. And the whole system needs to be verified well beyond component level testing.
“The sheer impossibility of using physical driving for that verification means a significant use of simulation is needed to explore the heavy tail of possible surprises – and to test the resulting system with different injection points.
“The Forrester report commissioned by Arm rightly recognises the huge size of the technical effort needed for this effort. It’s clear that limiting the operational design domain by bringing the technology to market in specific routes, in specific cities and potentially in specific conditions is one of the keys to making the problem tractable. And the simpler we can make the initial segues to market, the sooner we will see this technology deployed.
“Most firms working on the complexity of the software, simulation and verification are seeking out partners to ensure the work can be shared and speeded up. FiveAI is no exception.
“With that and with contained operational design domains, 10 years to get to market without safety drivers is too pessimistic for these early level 4 deployments.
“A concentrated effort we and others will apply to get systems launched in the next 4-5 years will yield a huge societal gain as we unlock the means to deliver shared urban on-demand mobility. And that’s transformative for our cities – for everyone.”
The Arm-Forrester survey of 54 automotive industry experts shows the biggest hurdle in turning prototype autonomous vehicles into mass market models is how to cut production costs, but the area energising engineers most is how to tackle functional safety and cybersecurity, according to Robert Day, director of automotive solutions & platforms at Arm.
He writes: “Technology challenges mean the race to the mass deployment of artificial intelligence-powered, self-driving cars will not see any companies breaking the speed limit.
“According to the experts surveyed, the current crop of prototype autonomous vehicles are at least 10 years away from mass deployment. While this may not be too surprising, it is the scale of the technology challenge that is startling.
“It signals a vital near-term shift from concept self-driving car technologies to functionally safe, secure, and affordable designs as autonomous vehicles evolve from prototype to production.
“As one US-based director of autonomous vehicle programs for an Asia Pacific carmaker put it: ‘It’s going to be a slow continuum of further capabilities.’ That is a true understatement of where the most advanced self-driving car systems are today in my opinion.
“The biggest challenge is around the foundational computing. All of my autonomous vehicle maker contacts tell me that the current banks of power-hungry computers fed by arrays of expensive sensors won’t scale.
“And despite the huge processing power deployed, these systems still require backup human drivers. The truth is the technology in test cars today must be replaced by trusted AV technology that is vastly more efficient, affordable and capable.
“It is actually a key opportunity for Arm as mindsets shift from concept to commercial and there’s a new focus on extending vehicle ranges through far more efficient AI-capable compute technology that is safe, secure and cost optimised.”
In the race to self-driving vehicle commercialisation, the top five areas of concern according to Forrester’s panel were: ‘creating a reliable system’, ‘manufacturing efficiency’, ‘functional safety’, ‘cyberthreats’ and ‘keeping costs low’.
An autonomous vehicle perception engineer from a US car company said: “The components have to be almost military-grade robust. It’s the part of the vehicle that’s not really where it needs to be, especially when it comes to reliability and fault tolerance.”
One engineer told Forrester he could see a path to 96-97 per cent of the way to full autonomy but the final 3-4 per cent would be exponentially harder to achieve. Given the technologies currently available, he couldn’t see how the ultimate goals of self-driving vehicles would be achievable.
Day added: “The biggest issue that particular engineer spoke about was the public’s apparent expectation that autonomous vehicles would be completely safe – perhaps 10x as good as the average human driver.
“This theoretically could be as far-fetched as dealing with a meteor falling from the sky. Or, more realistically, the issues that prototype autonomous cars are failing on today according to the latest State of California Department of Motor Vehicles test data which points at self-drive systems disengaging due to the ‘curvature of road’ and exhibiting ‘inappropriate trajectory and acceleration’.
“Working for Arm, I know that most (falling meteors aside) challenges are solvable and functional safety, security and compute advances that I see coming will lead to safe and secure Level 4 and 5 vehicles.
“During this technology evolution we’ll also save many lives, as we’re already seeing with leaps in mainstream Advanced Driver Assistance Systems (ADAS) – much of which is based on Arm-designed technologies.
“My conclusion from reading the Forrester report, and what I know, is that fully-autonomous vehicle fleets will happen. The deployment of the first truly self-driving cars are likely to be restricted, such as cars only operating in pre-defined (and well-mapped) areas, in certain environmental conditions (e.g. daytime, dry conditions) and at limited speeds. This is, by the letter of the SAE law, Level 4.
“Then, we’ll see a gradual roll-out as trust builds through compute advances and artificial intelligence becomes more ‘intelligent’. So, the question is not ‘if’ but ‘when’ on autonomous vehicle fleets – and that cuts to the core of the need now to shift from prototyping to trusted secure and safe market-capable technologies.”