Advanced driver-assistance systems are designed to improve safety, but their real-world behaviour can be very different from laboratory results. Thatcham

Research is now testing technologies such as automatic emergency braking, lane keeping and intelligent speed assistance on public roads as well as closed tracks. Its aim is to measure not only whether these systems work, but whether drivers can realistically live with them every day.

Real Roads Reveal New Problems

At Thatcham’s Gamston testing facility, engineers have found surprising differences between vehicles.

During one test, a Tesla Model Y travelling at 12km/h automatically stopped before a simulated vehicle. At 6km/h, however, the system did not intervene and the driver had to brake manually.

A BYD Dolphin Surf stopped automatically at just 4km/h, but then began moving again a few seconds later. In real traffic, such behaviour could confuse a driver who believes the hazard has already been handled.

These examples highlight why controlled regulatory testing cannot reveal every possible interaction between assistance technology, drivers and unpredictable road conditions.

Accuracy Can Look Better Than It Is

Thatcham’s engineers are particularly interested in intelligent speed assistance, which detects speed limits and warns drivers when they exceed them.

Traditional testing measures accuracy across the overall distance travelled. Thatcham instead checks performance every time the speed limit changes.

The difference can be substantial. An MG ZS achieved 91.3% accuracy using the distance-based method, but only 74.3% when individual events were assessed. That means it displayed an incorrect limit in roughly one out of four situations.

A BMW i5 performed better, scoring 98.39% by distance and 90.3% in event-based testing.

Principal ADAS engineer Yousif Al-Ani explained that even the strongest result still meant the system was wrong about once in every ten events. In his view, repeated mistakes can undermine trust and eventually persuade drivers to disable technology intended to protect them.

False Braking Can Frustrate Drivers

The programme also tracks unnecessary interventions. During testing, the Tesla Model Y recorded 122 instances of unnecessary braking, compared with eight for the BMW i5 and six for the MG ZS.

The Tesla often responded earlier to potential hazards, which can provide useful protection, but greater sensitivity also produced more unwanted reactions. Thatcham believes an assistance system that regularly surprises or irritates its driver risks being switched off altogether.

Lane-keeping performance was generally more encouraging. The BMW i5 recorded no false interventions in the measured tests, while the MG ZS registered only 0.43 per 100km.

More Customisation Could Help

Thatcham argues that drivers should have greater control over how assistance systems behave. Rather than forcing motorists to choose between maximum intervention and switching a feature off completely, manufacturers could offer adjustable sensitivity settings that remain saved after the vehicle is restarted.

Al-Ani noted that BMW has made progress in this area by allowing drivers to configure assistance features according to their preferences.

The wider goal is straightforward: ADAS should not merely satisfy regulations. It needs to work reliably in ordinary traffic, communicate clearly and cooperate with the person behind the wheel. If manufacturers can achieve that balance, drivers may be much more willing to leave these safety systems switched on.