Driverless cars are now tested on our roads, so should we be worried?


Updated

March 23, 2018 06:32:59

Driverless vehicles are already being tested on Australian roads, but in the wake of the first pedestrian death in a collision with a self-driving test vehicle in the US, experts are asking whether the right safeguards are in place.

In order for autonomous vehicles to be as safe as possible on the road with other vehicles and around pedestrians, they have to be tested in urban environments.

But using public streets and other road users as a test environment raises complex issues.

“Not all companies are ready,” said associate professor Hussein Dia from the Smart Cities Research Institute at Swinburne University of Technology.

The issue has come to the fore after a woman who was crossing a road at night was struck and killed by a self-driving car that Uber was using for test drives on public roads in Tempe, Arizona.

Local police have released video recordings from the car’s onboard cameras showing the moments just before 49-year-old Elaine Herzberg was hit.

The video filmed through the windscreen shows the woman appearing suddenly from the dark as she crossed right in front of the car.

The vehicle does not appear to respond at all.

Computing power

All autonomous cars have cameras and sensors that collect data about a car’s environment, but it is the software in the onboard computer that processes the information and makes decisions about what action the car should take.

In Dr Dia’s view, the quality of a driverless car’s software is crucial to its safety, and some companies’ software performs better than others.

“Some companies have been testing the self-driving software for years and their algorithms are much more developed than others,” he said.

“There needs to be more scrutiny of the underlying AI [artificial intelligence] systems before the autonomous vehicles are allowed on open roads.”

Dr Zubair Baig, a senior lecturer in cyber security at Edith Cowan University, would like greater involvement in the testing process by local governments that allow trials on their public roads.

He notes that the systems that control self-driving cars have to be tested in real-life scenarios, under varying conditions, in the same way that aircraft are tested in flight before they are approved for passenger transport.

“A long checklist must be ticked before an autonomous vehicle is allowed onto real roads for testing, to ensure that fatalities in such cases are non-existent,” he said.

As the age of the driverless car swiftly approaches, testing is taking place at facilities across the world as carmakers and software companies race to get the best autonomous car on the road.

It is highly competitive, and highly secretive.

One suggestion for making tests on public roads safer is to compel companies to lift their virtual bonnets, and reveal some of the secrets of their tightly guarded algorithms.

“Regulators need to ask developers to reveal their results in certain environments before they are allowed to test on the roads,” Dr Dia said.

The future is now

Cars with driverless features are already on the road in Australia, but fully autonomous cars are still a short way off.

Perth is set to test on-demand driverless cars on private roads later this year, with a view to them operating on city streets later in the year.

Victoria, New South Wales and Queensland have all made it possible for autonomous vehicles to be tested on their roads, but Australia’s regulators have been taking a cautious approach, so trials of these cars have not yet reached the same stage they have overseas.

Tempe, where the fatal collision between the Uber and the pedestrian happened, is part of greater Phoenix, which the Governor of Arizona and local business groups have been promoting as a centre for the driverless car industry.

Uber was testing a fleet of driverless cars there at the time of the accident, but has now announced it had suspended all tests of autonomous cars on public roads in Pittsburgh, San Francisco, Toronto and Phoenix.

Critics have said states like Arizona have been too quick to allow trials on their roads.

Only car’s data, central servers will reveal what went wrong

As policy makers examine the broader implications of Ms Herzberg’s death, investigators are examining the finer details of the moment she was hit.

Soon after the incident, police said she tried to cross the road without using a pedestrian crossing, but Dr Dia said any self-driving car should be able to detect a pedestrian, even if they were not on a crossing.

“I would have expected the self-driving software to respond,” he said.

Dr Baig is not so sure.

“Robots are able to make judgements, but they can’t always predict human behaviour with perfect accuracy,” he said.

As Dr Baig sees it, a combination of factors may have led to the vehicle not detecting the pedestrian, and only data from the vehicle and the central servers will reveal what went wrong.

Dr Dia agrees that the video alone cannot explain why the car didn’t detect the pedestrian, and that investigators will need to look at data from the car’s sensors and the vehicle’s real-time analytics about the surrounding environment.

“The investigators would now be working to determine what other sensors were onboard, what data they were providing and how the self-driving software was analysing the data,” he said.

But there is a way the accident could have been avoided.

“The operator could have possibly been notified by the vehicle to take charge,” Dr Baig said.

What the video released by Tempe police shows from inside the car is that the safety driver was looking down inside the vehicle for several seconds before the collision, and appears to have had no warning about a person on the road.

“During test runs, the purpose of having an operator in the vehicle is to intervene when such hazards present,” Dr Baig said.

“Video footage does indeed reveal a shocked look on the face of the operator right before the vehicle hit the pedestrian.”

Carmakers confident

Mercedes Benz said the automated technology in their cars sold in Australia was so rigorously tested, they could often respond better than humans.

“The amount of data that is gathered from the testing of autonomous vehicles is more than the data that Airbus use to test the A380,” Mercedes Benz public relations manager David McCarthy said.

He said self-driving cars had multiple cameras and sensors that could detect things that could not be seen with a single pair of human eyes, and could therefore activate controls to avoid a situation that a human might not.

Dr Dia said of the hundreds of thousands of people killed on the road every year, 90 per cent of fatalities were caused by human error, and that by removing humans from driving cars, that death toll could be drastically reduced.

“They still will make mistakes, but they will be better than humans,” he said.

This raises another significant ethical dilemma — should we allow self-driving cars on the road when they are 90 per cent ready, or should we delay their adoption to improve them further, knowing during that time more lives will be lost due to human error?

Topics:

computers-and-technology,

science-and-technology,

robots-and-artificial-intelligence,

automotive,

australia,

united-states

First posted

March 23, 2018 06:16:15



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *