Just a few years ago, futurists painted pictures of a world without drivers: cars taking children to school on their own, taxis arriving on call without a human behind the wheel, and accidents being reduced to statistical zero. The reality of 2025 is less glossy: there has been no massive rollout of autonomous vehicles, and the prospects look much more modest than in Tesla or Waymo adverts.
Technology has hit a complexity ceiling
The problem is not in loud promises, but in the reality of complex urban environments. AV sensors (LiDAR, radar, cameras) combine to create the car's "eyes", but even they can go blind in rain, snow, or fog. And neural networks that control cars still get lost in unusual situations: unexpected pedestrians, roadworks, sudden accidents ahead - all this is a "long tail" of scenarios that cannot be fully predicted.
Go Deeper:
The SAE classification system defines 5 levels of car autonomy: Level 0 - no automation at all, the driver controls everything. Level 1 - assistance in certain functions (cruise control, lane keeping). Level 2 - partial automation, the car controls the speed and direction, but the driver must be alert (Tesla Autopilot). Level 3 - conditional autonomy, the car drives itself in certain conditions, but the driver must respond to system requests. Level 4 - high autonomy in designated areas (robotaxis), without human intervention, and with restrictions on territory and weather. Level 5 - full autonomy anywhere and anytime, no steering wheel or pedals required. Today, Levels 2-3 are available in mass-produced cars, and Level 5 remains a distant goal.
Public trust is under the brakes
High-profile accidents involving drones(Cruise in California or Uber in Arizona) have left a mark on people's minds. An AAA survey in early 2025 showed that 60% of American drivers are afraid to get into an autonomous car. Scepticism is also fuelled by marketing gimmicks, when L2 systems are presented as "full-fledged autopilot".
Legislation and money are two more brakes
While regulation in the US is fragmented between states, and a unified framework is only being prepared in the EU, China is forging ahead with centralised programmes. Manufacturers are afraid of liability: who is to blame if a drone causes an accident - the driver, the manufacturer, or the software developer?
Expensive toys for the rich?
LiDAR is getting cheaper, but a full set of sensors and computers for an autonomous car still adds tens of thousands of dollars to the price of a car. Mass production should bring prices down, but without stable demand, it's a vicious circle.
Cyber threats and privacy - new challenges
Hacking into a car via Wi-Fi or spoofing a GPS signal? This is not a scenario from a TV series, but real cases ( 1, 2 more) by researchers. And while manufacturers are looking for a balance between data collection for AI training and user privacy, data leakage scandals do not add to trust.
Bottom line: not a sprint, but a marathon
Autonomous cars have not disappeared from the radar - they are already operating in the form of robotaxis in the US and China, autonomous trucks on the highways and shuttles on campuses. But the road to making them commonplace on the roads is a decades-long marathon. We need to synchronise progress in technology, law, economics, and cultural perception.
And while we are waiting for Level 5 autonomy in every yard, it is worth remembering that even a human driver is not always perfect, but we are not yet ready to hand over the wheel to algorithms without the conditioned reflex "keep your hands on the wheel". This is a race where the main thing is not speed, but reliability.
For those who want to know more
Han anslöt sig till gg-teamet 2011. Henri är en man med varierande intressen och skriver om prylar och militär utrustning. Han förstår sig på konsumentelektronik och har stor erfarenhet av olika typer av spelutrustning.
