Who's in control? More transparency in driver-assist tech required

If you have driven a vehicle manufactured in the last few years, you would be aware of the myriad advancements in driver assist technology that are becoming ubiquitous in contemporary cars.

Radar cruise control, lane keep assist, automatic blind spot monitoring and autonomous emergency braking are just some of the technology-infused features that are helping drivers stay safe on the road day-to-day.

Voice activated commands for entertainment, air conditioning and navigation add to the comfort for everyone in the vehicle. The introduction of Apple CarPlay and Android Auto lets you use compatible apps directly from your car's inbuilt infotainment system, ensuring quick access to music or podcasts so those long, cumbersome highway drives are more tolerable than ever.

But for all the breakthroughs and innovation, there is a sentiment that the technology is making today's drivers complacent.

Genuine self-driving tech available at scale to consumers around the globe is mostly in the realm of science-fiction, at least for the time being. Even by more optimistic projections, large scale rollout of Waymo-esque technology is years away.

Until then, vehicle manufacturers and end users share responsibility for the safest possible operation of the increasing amount of technology we see in cars in 2026, said PhD candidate at Macquarie University, George Nasser.

"It's one of the most important questions because it comes back to this idea of who's in control," Nasser said.

Drivers need to understand what a vehicle can do and where its limitations lie.

The challenge is particularly important because drivers can develop inaccurate mental models of vehicle capabilities, influenced by marketing terminology, online demonstrations and their own experiences with automated features.

Under the commonly used vehicle automation framework, levels range from zero, where no driving automation is active, to level five, representing full automation across driving conditions.

Most new mass market vehicles operate at levels one or two, where systems can assist with functions such as steering, acceleration and braking but drivers remain fully responsible.

Level three systems introduce conditional automation, allowing drivers to disengage under very specific conditions, although they must still be able to respond when the system requests intervention.

The distinction between these levels is critical to ensuring drivers understand their responsibilities.

Even when a vehicle handles much of the driving, drivers can struggle to resume control quickly because they are no longer continuously engaged, Nasser explained.

"When you're in an automated state, even though the system is assuming responsibility, you're no longer in physical control, so your ability to get back in control is diminished," he said.

Drivers typically have approximately five to seven seconds to recognise a system limitation and take corrective action in relevant situations.

However, a driver who has become disengaged may need longer to understand what is happening and respond appropriately.

Nasser cited the fatal Uber autonomous vehicle crash in Tempe, Arizona in 2018 as an example of the risks associated with drivers failing to respond to system limitations. The incident involved a test vehicle that struck a pedestrian while operating in autonomous mode.

Improving communication between drivers and vehicles

Nasser's PhD research is examining how automated driving systems can help drivers remain engaged and better understand a vehicle's capabilities.

Instead of relying on drivers to remember training or interpret basic warnings, vehicles should communicate information about their confidence and limitations in real time.

A recent experiment tested a feature designed to provide drivers with additional information about an automated system's confidence in its ability to perform the driving task.

Rather than indicating that driver assistance was switched on or off, the feature introduced an intermediate state informing drivers when the system's confidence had decreased.

Drivers receiving this additional information performed smoother takeovers than those who did not receive the cue.

They were better able to maintain their vehicle's position in the centre of the lane when approaching a kerb, suggesting the additional information helped them anticipate the need to intervene.

The findings suggest that improving the transparency of automated systems could help people recognise when they need to resume control.

"If we provide drivers transparency as opposed to just saying assistance on, assistance off, that allows them to anticipate what the system is going to do and react better," Nasser said.

The research is particularly relevant during those long highway drivers, where drivers may become accustomed to automated functions such as adaptive cruise control and lane-keeping assistance.

A driver travelling between Sydney and Canberra with their partner and children in the car, for example, might spend an extended period relying on these functions while listening to a podcast or participating in a conversation.

Over time, vigilance could decline, leaving the driver less prepared to respond when circumstances exceed the system's capabilities.

But a warning that the system's confidence had decreased could help bring the driver back into full concentration before an intervention actually became necessary.

Nasser said the objective is to establish a more effective relationship between the driver and the vehicle, in which people continuously understand what the system is doing and what it can and can't handle.

Training, regulatory challenges persist

Conventional driver training is not going to provide a complete solution, given the difficulty of requiring every customer to complete extensive simulator sessions before using an advanced vehicle.

Even when training was provided, drivers could forget the information over time or become accustomed to the system's behaviour.

Nasser argued that transparency needs to be incorporated into vehicle design, allowing drivers to understand system limitations without relying on previously learned information.

However, designing effective warnings presents its own challenges. Vehicle cabins already contain numerous alerts, sounds and other prompts, meaning additional information must be communicated without overwhelming drivers.

Manufacturers' goal should be to develop intuitive and visually minimal cues while still providing enough information to support safe decisions.

He also called for greater scrutiny of how manufacturers market automated driving technologies.

Terms such as Tesla's Autopilot can influence expectations about a vehicle's capabilities, potentially creating a mismatch between what drivers believe a system can do and what it is actually designed to handle.

Manufacturers have a vested interest in promoting their technology, making appropriate regulation important to ensure marketing language accurately reflects system capabilities.

"We need to be very careful with our terminology and ensure that it's aligning with what it's capable of, and not creating broader misconceptions," Nasser said.