Autonomous vehicles are already far more capable than many people realize, but they're also less capable than many assume, according to researchers who spent over 100 days testing a self-driving car on Australian roads. The study identified more than 500 safety-critical events where the system required driver intervention, but the findings suggest the real bottleneck isn't the technology itself, it's the roads the vehicles are trying to navigate.

The research team, which tested a Tesla Model Y with Full Self-Driving (FSD) on Queensland roads, discovered that most problems weren't rare, unpredictable situations. Instead, they were everyday scenarios that human drivers handle almost automatically. This reframes the entire conversation about autonomous vehicle readiness, shifting focus from pure technological advancement to infrastructure adaptation.

What Specific Road Conditions Confuse Self-Driving Cars?

The testing revealed a pattern of recurring challenges that have little to do with cutting-edge AI limitations. A small neighborhood bridge that most drivers barely notice became a consistent problem point. Almost every time the autonomous vehicle approached it, the system appeared confused and began weaving from side to side. This pointed to a fundamental issue: road markings and layouts that are intuitive for humans can still confuse autonomous driving systems.

Time-restricted speed limits, such as school zones, caused similar problems. Researchers had to intervene more than 90% of the time in these scenarios. In one notable case, the vehicle obeyed a school-zone speed limit correctly, except it was evening, long after school hours, demonstrating how the system struggles with contextual understanding.

Beyond speed zones, the testing identified several other recurring issues:

• Railway crossings and boom gates: In one incident, the car ahead stopped near a railway crossing, but the autonomous vehicle continued moving and could have ended up stopped on the tracks, requiring emergency driver intervention.
• Complex roundabouts: The vehicle frequently struggled at roundabouts, which are particularly common in Australia, indicating difficulty with multi-directional traffic flow.
• Poorly marked streets: Steep streets crowded with parked cars and unclear lane markings consistently caused the system to misread the driving environment.
• Zipper merges: Australia's zipper merge rule, which relies on subtle human judgment and informal negotiation, proved challenging; in one incident, neither the autonomous vehicle nor the other vehicle slowed down at the merge point.
• Weather conditions: Harsh or variable weather that obscured lane markings and road edges significantly reduced the system's accuracy.

Perhaps most striking: in more than 100 days of daily testing, the researchers did not complete a single trip in which FSD drove independently from start to finish without requiring some form of intervention.

How to Make Roads More Autonomous-Vehicle Friendly

The researchers propose that the solution isn't simply waiting for smarter vehicles, but rather meeting in the middle by improving both vehicle intelligence and infrastructure simultaneously. This doesn't require expensive technological overhauls or complete road reconstruction. Many fixes are straightforward and can be implemented during ordinary public works and maintenance.

• Clearer lane markings: Repeating lane markings more frequently gives automated systems more opportunities to recognize them, correct misreadings, and adjust safely.
• More consistent signage: Instead of a single speed-limit sign or a bicycle symbol painted only at the start and end of a lane, agencies could repeat these cues more often throughout the relevant area.
• Better-maintained road surfaces: Potholes and damaged pavement contribute to confusion; regular maintenance improves both human and machine readability.
• Less ambiguous intersection design: Clearer visual cues at intersections reduce interpretation errors for autonomous systems.
• Reliable speed-limit information: Consistent, clearly marked speed-limit data helps systems understand context-dependent limits like school zones.

The researchers also suggest that autonomous vehicles themselves can become part of the solution. As they travel, these vehicles can act as mobile sensors within the road network, flagging potholes, faded markings, damaged signs, and bottlenecks. This real-time data could help agencies detect problems earlier and prioritize maintenance more effectively. Additionally, data from autonomous vehicles could support congestion mitigation, incident detection, and traffic management, transforming them from passive users of the road network into active participants in its improvement.

What Does This Mean for the Robotaxi Industry?

While the Australian testing focused on Tesla's FSD system, the findings have broader implications for the robotaxi industry, including companies like Waymo and Zoox that are expanding operations globally. The research suggests that regulatory approval and technological capability are only part of the equation. Infrastructure readiness is equally critical.

Zoox, Amazon's autonomous vehicle subsidiary, recently unveiled an updated version of its purpose-built robotaxi with significant rider-focused improvements. The company incorporated feedback from half a million riders during testing and early deployments, adding ergonomic enhancements like better seat padding, larger cupholders, improved phone charging pads, and two-way audio communication with support staff. These upgrades reflect a shift in the industry's priorities: the early race focused on whether autonomous vehicles could safely drive through cities, but now companies must also prove that people will actually enjoy using them.

Zoox's vehicle still has no steering wheel or pedals and can drive bidirectionally, but the company made exterior changes focused on visibility and communication. Relocated bidirectional reflectors now rotate color to clearly distinguish the robotaxi's front from its rear, helping pedestrians and other drivers understand which direction it's moving. However, broader commercial deployment still depends on regulatory approval, with Zoox having petitioned the National Highway Traffic Safety Administration (NHTSA) for temporary exemptions from certain federal motor vehicle safety standards.

The Australian research suggests that as robotaxis scale beyond early testers to regular riders, the quality of the entire experience becomes critical. A stiff seat, a sliding phone, or poor communication with support staff can quickly turn a futuristic ride into an annoying one. But equally important is the infrastructure these vehicles operate within. Without clearer road markings, more consistent signage, and better-maintained surfaces, even the most advanced autonomous systems will struggle with everyday driving tasks that human drivers handle almost unconsciously.

For governments and the automotive industry, the implication is clear: preparing for autonomous vehicles requires coordination on multiple fronts. Major events like Brisbane's 2032 Olympic and Paralympic Games present opportunities for infrastructure improvements that could benefit autonomous vehicle operations while also improving overall road quality for all users.