Robotaxis could reach Australia by 2028, but hurdles remain

Australian streets could see commercially operated driverless taxis as early as 2028, according to an RMIT academic, but regulatory readiness, local safety testing and the economics of operating autonomous fleets will determine how quickly the technology can be fully implemented.

Overseas markets have already seen pilots and even full-scale adoption of self-driving transport systems.

US company Waymo has successfully implemented fleets of self-driving vehicles carrying passengers in San Francisco, California and Phoenix, Arizona, as part of its network of 10 markets. Trials are underway for similar services in the UK and across Europe.

Chinese AI and technology firm Baidu has deployed a self-driving taxi service in Wuhan with over 1000 vehicles in its growing fleet.

Reza Hoseinnezhad, professor in the School of Engineering, RMIT University, said a limited robo-taxi service could become viable in Australia within the next 24 months, provided operators can secure the necessary approvals and demonstrate that their vehicles can operate safely in local conditions.

His forecast comes as governments and technology companies continue to assess how autonomous vehicles could be introduced to Australian roads.

Self-driving companies such as Waymo and Tesla will require a clear pathway covering vehicle approval, accountability, insurance and accident response and liability.

Australia's regulatory structure presents an additional challenge because vehicle regulation occurs at the national level, while road rules and other requirements can vary between states.

"Australia's combination of national vehicle regulation and state road rules makes coordination particularly important," Hoseinnezhad said.

He argued that safety cannot simply be treated as a box that has already been ticked, with regulators needing evidence that a particular autonomous driving system can operate safely in the Australian environment.

That evidence would also need to extend beyond an initial approval, with regulators requiring the ability to monitor systems as software is updated.

Investment is another consideration, particularly around local validation, fleet maintenance and incident response.

However, a predictable regulatory process could itself encourage greater investment.

The nation's most pressing challenge will be establishing a credible pathway from autonomous vehicle trials to an accountable public transport service.

Self-driving models require local validation

Public confidence will also be an important factor as companies seek to deploy autonomous vehicles without a human behind the wheel.

Concerns about self-driving technology are understandable, particularly because autonomous systems can make unfamiliar mistakes and responsibility can become difficult to determine when something goes wrong.

A software defect could also potentially affect an entire fleet, rather than a single vehicle.

Hoseinnezhad cautioned against treating self-driving cars as a single category that can be given a blanket safety assessment.

Instead, each system needs to be assessed according to its capabilities and the conditions in which it operates.

He cited research from driverless services, particularly Waymo, as encouraging, noting that some analysis has found lower injury crash rates than comparable human-driving benchmarks.

Those findings should be considered within their limitations, Hoseinnezhad clarified, with much of the data produced by a commercial company with a vested interest, and performance demonstrated in a US city not automatically establishing safety in Australian conditions.

The relevant comparison should be made against human performance under comparable conditions, including risks to pedestrians and cyclists.

Individual failures should also continue to be investigated even when the overall safety record of an autonomous system appears favourable.

"Public confidence should be earned through evidence, and people should be able to scrutinise (self-driving companies)," Hoseinnezhad said.

That sentiment was shared by Hugues Blache, Associate Lecturer & Research Associate in the School of Civil & Environmental Engineering at UNSW, whose research has focused on the safety of self-driving systems.

"You can have the most efficient systems with a great research base or engineering developments, but if you don't have any safety controls behind it, this will not be acceptable," he said.

ANZ transport peak association Austroads is currently seeking tenders for the feasibility of self-driving trials.

While Hoseinnezhad considered putting in a bid, he ultimately decided against it because the maximum tender value was $200,000, which would not cover the cost of a university investigation.

With a lack of investment in the tender process, it would be difficult for Australian universities to justify using their resources on the proposal, he said.

All-camera vs multi-sensor debate

The debate over autonomous vehicle technology also extends to the sensors used to perceive road conditions.

Waymo's autonomous vehicles use multiple sensing technologies, including cameras, LIDAR and radar, while Tesla has pursued a camera-focused approach.

Camera-only autonomous driving is technically possible, with cameras offering advantages around cost, availability and the ability to interpret visual information, Hoseinnezhad said.

He distinguished between demonstrating that a system can drive using cameras and proving that it can do so reliably without a human fallback, however.

Cameras are particularly useful for recognising traffic signs, signals and road markings. LIDAR, meanwhile, can provide measurements of distance and three-dimensional structure, while radar can measure distance and relative speed and is generally more resilient in poor visibility.

Combining the technologies therefore provides a vehicle with different sources of information when one sensing system becomes unreliable.

"For a driverless service expected to work across varied conditions, my engineering preference would be complementary sensing," Hoseinnezhad said.

A system of additional sensors brings its own drawbacks, it should be noted, including higher costs, calibration requirements and greater software complexity.

LIDAR is susceptible to reduced effectiveness in rainy or foggy conditions, while multiple cameras can share vulnerabilities to glare, poor visibility or contamination.

One of the most important questions that must be answered for Australian conditions is how an autonomous vehicle behaves when its perception systems become unreliable, according to Hoseinnezhad.

"No sensor combination guarantees perfect judgement," he said.