Posted in todays West Australian.
Maybe somebody needs to tell them about a little ol WA company up the road that may have the solution.


Autonomous vehicles raise risk of cyber attacks

Olga de MoellerThe West Australian

Sat, 17 July 2021 6:23AM

Self-driving cars may remove the human error which plays a part in most crashes but research shows their control systems can be sabotaged with potentially disastrous outcomes for other road users.

It’s part of a wider issue with modern vehicles that rely on sophisticated driver assistance programs, which can be hacked to exploit vulnerabilities and play havoc on software that controls everything from windscreen wipers to critical components like the steering wheel, engine, transmission and brakes.

The risk was demonstrated in 2015 by the now-infamous “Jeep hack” by a couple of US researchers using wi-fi connection and a Cherokee’s multimedia system controller.

Fiat Chrysler subsequently recalled 1.4 million vehicles, including Rams, Dodges and Jeeps, for a software fix.

Nothing is safe these days — even music downloads with malware codes can get into your car’s infotainment system and infiltrate everything, which means your connected smartphone with your credit card information and passwords to bank accounts is fair game.

Autonomous vehicles (AVs) are particularly at risk, with Macquarie University’s department of computing identifying key vulnerabilities that make them susceptible to malware attacks — and how car makers can detect and prevent them.

Yao Deng Credit: Supplied/Supplied

PhD candidate at the university Yao Deng is investigating weak spots in a critical part of an AV’s computer vision system called the “convolutional neural network”, or CNN, used by robots and autonomous vehicles to recognise and classify images.

“There are many different kinds of attacks that can occur on self-driving or autonomous vehicles,” he said.

The CNN is the “brain” of an autonomous vehicle, with Mr Deng’s research investigating End-to-End driving systems which make decisions about speed and steering angle using cameras and laser sensors called LiDARs positioned on the car.

Other autonomous systems use CNN only to recognise the driving environment and can also be attacked, affecting the vehicle’s final decision.

“Unfortunately, CNNs can be easily fooled ... to make completely wrong decisions,” Mr Deng said, citing a recent example of a demonstration in the US in which Tesla autopilot was tricked into turning on rain wipers when there was no rain by an “adversarial attack” which made minute pixel changes to images scanned by the car’s cameras and LiDARs which were imperceptible to human eyes.

UWA’s AV bus. Credit: Supplied/Supplied

Such attacks meant autonomous vehicles could easily cause traffic accidents and jeopardise personal safety — and while work was under way around the world to protect autopilot systems, Mr Deng said current security measures often did not address inherent weaknesses in CNN logic.

Adding to the problem, CNNs were susceptible to malware attacks when the vehicle connected to the internet to upgrade software and firmware — and this malware could then intercept the images the vehicle received, falsifying the information sent to the computer.

“The biggest problem we face with neural networks is that we actually don’t know how they work, so we cannot interpret how they make decisions or make predictions — and whether these will be correct if the system is attacked,” Mr Deng said.

“This means that the deep learning system that drives the AV is its biggest vulnerability because a neural network is not like a normal program for which we can define the logic; it has so much data that it trains itself, just like a human, so we’re talking about artificial intelligence.

“We do the initial programming but we don’t know how the system learns the patterns or features to do a task.

“That’s why safety is such a priority for us. At this stage there have been instances of AVs making mistakes in the real world, but I am not aware of any malware attacks — yet.”

Mr Deng is testing various autonomous systems for bugs as part of his research and has recently collaborated with major manufacturers and computer scientists from Harvard, the University of California, Los Angeles, and the University of Sydney to identify CNN security threats.

University of WA robotics professor Thomas Braunl is leading the “nUWAy” autonomous shuttle bus project which has just started doing a 500m circuit on campus from the Reid Library to the business school.

Professor Thomas Braunl. Credit: Simon Santi/The West Australian, Simon Santi

Professor Braunl said malware was a broader problem which could affect all cars today and was not specific to autonomous driving.

“It could just be an error in programming that may be exploited at a later stage,” he said. “Your PC is updated daily to fix these problems but, with the exception of Tesla, most car companies take forever to fix bugs.”

He said the campus bus didn’t use neural networks for its driving operation.

“We have some (CNN) modules for detection of pedestrians ... and we’ll probably also use them as well for future extensions,” he said.

“But we did the software entirely ourselves and the bus can drive without any internet connection, though we usually have it on so we can monitor things.”

Australia on track

Australia is already laying the groundwork for autonomous vehicles in the near future, with initiatives including the RAC Intellibus, the UWA “nUWAy” shuttle bus, the Transport for NSW driverless shuttle bus pilot and the Coffs Harbour City Council Busbot.

Several Australian mining companies routinely use self-driving vehicles within closed sites and various government transport plans, such as Victoria’s North East Link road project and the Transport for NSW strategy, now include provision for AVs.

So far, most jurisdictions do not permit AVs to operate on public roads without a human in the driver’s seat, ready to take over at any moment.

But as AV manufacturers perfect their vehicle control systems, we are likely to see a shift where autopilot mode becomes far more acceptable — and eventually human drivers may be redundant.

The most recent KPMG global Autonomous Vehicles Readiness index rates Australia 15th in the world in its progress towards a self-driving future — and surprisingly ahead of China in 20th spot.

Singapore ranks first, just ahead of the Netherlands (which was number one in 2019), followed by Norway, the US and Finland making up the top five.