The promise of fully autonomous cars that will free us from the need to touch the steering wheel has been with us for over a decade. Although recent years have brought spectacular development in robotaxis in the USA and Asia, the widespread revolution still faces powerful resistance – technological, legal, and social. Where exactly are we today, and will we ever see the era of driverless vehicles?

Introduction: Anatomy of an unfulfilled promise

When, in the middle of the last decade, leaders of the tech and automotive industries competed in declarations that by 2020 millions of fully autonomous cars would hit the roads, the world believed in this utopia. Visionaries like Elon Musk promised the imminent arrival of an era where the steering wheel would become a relic of the past, and traveling by car would resemble relaxing in a living room. Reality, however, turned out to be much more complicated. Today, in the second half of the 2020s, it is clear that the road to full autonomy is not a short sprint, but an exhausting marathon. Although technology has made giant strides, the collision with physical, legal, and social realities has forced many players to revise their plans. In this context, it is worth remembering that artificial intelligence does not ask if we are ready – it simply enters subsequent spheres of our lives mercilessly, redefining transport as we know it.

Levels of autonomy according to SAE J3016: Where does the line really lie?

To talk reliably about autonomous vehicles, we must refer to the widely accepted standard created by SAE International (Society of Automotive Engineers). The J3016 standard defines six levels of driving automation, from no support at all to full machine independence. Understanding these differences allows us to separate marketing noise from facts.

1. Level 0 (No Automation): The driver performs all driving tasks, even if the car has warning systems (e.g., blind-spot sensors).
2. Level 1 (Driver Assistance): The vehicle can control either speed (adaptive cruise control) or steering (lane-keeping assist). The driver is constantly in control of the situation.
3. Level 2 (Partial Automation): Systems can control both speed and steering simultaneously. Examples include Tesla Autopilot or GM Super Cruise. Although these systems seem advanced, the driver must constantly monitor the surroundings and keep their hands on the wheel (or be ready to take control immediately).
4. Level 3 (Conditional Automation): This is a real turning point. Under specific conditions (e.g., traffic jams on a highway up to a certain speed), the vehicle takes full control, and the driver can take their eyes off the road. However, they must be ready for the system's request to take over the steering within seconds. An example of such an implementation is the Drive Pilot system from Mercedes-Benz, certified in selected markets in Germany and the USA.
5. Level 4 (High Automation): The vehicle drives completely on its own in designated zones and conditions (the so-called Operational Design Domain – ODD). In case of problems, the car can safely pull over to the side of the road, even if the driver does not react. This is the level at which today's robotaxis, such as Waymo, operate.
6. Level 5 (Full Automation): The Holy Grail of engineering. The vehicle handles any conditions, on any road, in any weather, without the need for a driver or even the installation of a steering wheel and pedals. Currently, Level 5 remains solely in the realm of theory and long-term research plans.

Years ago, as perfectly illustrated by an archival conversation with Dr. Aleksandra Przegalińska about artificial intelligence and autonomous cars, experts warned that moving from Level 2 to Levels 4 and 5 would require not just better computers, but a complete change of the technological paradigm.

Technological breakthroughs of recent years (2024-2026)

The last two years have brought fundamental changes in how engineers approach the problem of autonomous driving. First and foremost, a revolution has taken place in the area of artificial intelligence algorithms. The traditional approach based on rigid programming rules (so-called rule-based systems), which tried to predict every possible situation on the road, is giving way to an end-to-end deep learning approach. In this model, neural networks are trained on hundreds of thousands of hours of video footage and sensor data. The system learns to map camera images directly to steering, throttle, and brake inputs. This allows for much smoother, more natural driving that resembles human driver behavior.

Another area of dynamic development is sensor technology. For years, the justification for using LiDAR (Light Detection and Ranging) sensors was a bone of contention in the industry. While some manufacturers stubbornly claimed that cameras alone were enough for autonomy (the vision-only approach), most market leaders opted for sensor fusion. In recent years, the prices of LiDARs have dropped drastically, and their design has evolved toward solid-state technology (devoid of moving mechanical parts), which has drastically increased their reliability and facilitated integration with the bodies of production vehicles. Modern systems combine the precision of LiDARs with the range of modern imaging radars and high-resolution cameras, creating a three-dimensional, constantly updated model of the vehicle's surroundings in real-time. High-definition (HD) maps have also become an essential piece of the puzzle. These are not ordinary navigation maps, but centimeter-accurate digital twins of roads, containing information about curb heights, traffic light locations, road gradients, or permanent road signs. An autonomous vehicle constantly compares what its sensors see with the HD map, which allows it to instantly localize itself in space with an accuracy of a few millimeters.

The software war: Vision vs. sensor fusion

In the world of autonomous engineering, a deep ideological divide has emerged, symbolized by the dispute between Tesla and the rest of the industry. For years, Elon Musk has consistently pushed a vision-only approach (Tesla Vision), arguing that since humans navigate roads using only their eyes (biological cameras) and their brain (a biological processor), artificial intelligence should do exactly the same. Tesla has even abandoned traditional ultrasonic radars, basing its Full Self-Driving (FSD) systems on advanced video image analysis using neural networks. The advantage of this approach is a drastic reduction in vehicle production costs – cameras are cheap, easy to install, and do not spoil the car's aerodynamics. Critics, however, point out that cameras, like human eyes, are extremely susceptible to being blinded by the sun, dirt, snow, or sudden changes in lighting. The lack of a physical sensor measuring distance (such as radar or LiDAR) means the system must estimate depth based on a two-dimensional image, which in extreme cases can lead to tragic interpretative errors.

Waymo and most traditional automotive manufacturers follow a completely different philosophy. In their view, safety requires redundancy – that is, duplicating safety systems. If one sensor fails or is blinded, another must immediately take over its function. This is why Waymo vehicles are packed with technology: they have several LiDARs (creating a precise 3D point cloud around the car), millimeter-wave radars (which perform well in fog and rain), and over a dozen high-resolution cameras. Such sensor fusion allows for the creation of a system that is extremely resistant to interference, but it comes at a huge price. Until recently, the cost of equipping a single Waymo test vehicle exceeded the value of the car itself, which effectively prevents the sale of such cars to individual customers. It is a technology designed for commercial fleets, where costs are amortized thanks to the vehicle's continuous operation as a taxi.

Leaders of the race: Who is really calling the shots?

The landscape of companies fighting for dominance in the autonomous transport sector has consolidated. Analyzing the market, it is worth looking at the profiles of leaders and visionaries of artificial intelligence who directly fund and design these breakthrough solutions. At the head of the peloton remains Waymo, a subsidiary of Alphabet (Google). Waymo is the only one successfully scaling its robotaxi services in demanding urban environments such as San Francisco, Phoenix, Los Angeles, and Austin. Their fleet performs thousands of fully autonomous commercial trips daily, proving that Level 4 is technologically achievable and safe.

Cruise, backed by General Motors, has gone down a completely different path. After a spectacular image and operational crisis at the end of 2023, when the California regulator revoked the company's license following an accident involving a pedestrian in San Francisco, Cruise had to completely halt operations and reorganize its safety procedures. The company's return to the roads with safety drivers in 2024 and 2025 showed how fragile public trust is and how rigorous standards in this industry must be.

Alongside the giants of urban mobility, companies targeting autonomous trucking (hub-to-hub), such as Aurora Innovation, are developing dynamically. Founded by veterans of autonomous projects at Google, Tesla, and Uber, Aurora focuses on automating highway travel for massive Class 8 trucks. Their business model assumes the launch of permanent, driverless logistics corridors (e.g., in Texas), which could revolutionize supply chains struggling with a chronic shortage of professional drivers. The hardware and software foundation for many of these ventures is provided by Nvidia with its Drive Thor series of onboard supercomputers, capable of processing hundreds of trillions of operations per second.

Challenges on the road to widespread adoption

Despite undeniable successes, the widespread adoption of fully autonomous cars faces powerful barriers that cannot be solved by simply increasing processor computing power. We can divide them into four main categories:

• Technological challenges (edge cases): Machines handle ideal weather conditions in California or Arizona perfectly. However, heavy snowfall, torrential rain, mud covering sensors, or dense fog drastically limit sensor effectiveness. A separate problem is the so-called "edge cases" – extremely rare, unpredictable situations on the road, such as an unusually behaving pedestrian in a costume, an animal jumping out of the forest in an unexpected way, or manual traffic control signals given by a police officer. The human brain can instantly interpret the context of such events thanks to general intelligence and life experience; AI systems still have a huge problem with this.
• Legal and insurance barriers: Who bears responsibility for an accident caused by a Level 4 or 5 autonomous vehicle? The software manufacturer? The sensor supplier? The vehicle owner who was fast asleep in the back seat? The lack of uniform, international legal frameworks paralyzes market development. While in the USA regulations are decentralized and individual states (e.g., California, Texas) implement very liberal testing laws, the European Union approaches the topic with great reserve, prioritizing rigorous safety certifications and personal data protection.
• Road infrastructure: Autonomous vehicles need a clear environment. Faded lane markings, lack of vertical signs, unexpected detours with chaotic temporary signage – all this can confuse algorithms. The future requires the implementation of V2X (Vehicle-to-Everything) technology, i.e., a system of wireless communication between vehicles and urban infrastructure (e.g., smart traffic lights), which involves huge financial outlays by states and local governments.
• Social acceptance and ethical dilemmas: People show asymmetry in assessing errors. We are able to accept the fact that a drunk or tired human driver causes a tragedy, but an algorithm error triggers public outrage and demands for an immediate ban on tests. On top of this, there are classic, though sometimes exaggerated, ethical dilemmas (e.g., who the system should save in a no-win situation – pedestrians or passengers) and justified concerns about the loss of jobs for millions of professional drivers around the world.

Legal regulations in practice: Europe lagging, USA in the vanguard?

Differences in the pace of implementing autonomous technologies globally result not only from technological barriers but primarily from the legal philosophy of individual regions. The United States has adopted a highly experimental and decentralized approach. The National Highway Traffic Safety Administration (NHTSA) sets general safety frameworks, but individual states decide on allowing driverless vehicles on the road. California and Texas have become testing grounds where companies can test their prototypes in real urban traffic, often with minimal restrictions. Such a model fosters innovation but also raises serious controversies when incidents occur – such as robotaxis blocking fire trucks or hitting pedestrians. Every such incident triggers an immediate public reaction and the threat of license revocation, which shows that American liberalism has its limits.

The situation looks completely different in the European Union, where the precautionary principle dominates. European regulations, based on the work of the UN Economic Commission for Europe (UNECE), introduce autonomy technology extremely slowly, taking care of every legal detail. A breakthrough was the approval of Level 3 ALKS (Automated Lane Keeping System), but with rigorous speed limits (initially up to 60 km/h) and only on highway-type roads without pedestrian or bicycle traffic. In Poland, the legal framework for autonomous vehicles was outlined in the Road Traffic Act (Art. 65b et seq.), which enables research on autonomous vehicles on public roads. In practice, however, the procedures for obtaining appropriate permits, requirements regarding the presence of a safety driver, and the need for insurance for huge amounts make Polish roads almost free of such tests. For local engineers and startups, the barriers are not only financial but also bureaucratic, which pushes our market into the position of an observer of the global race.

The economics of autonomy – why it must pay off

The ultimate validator of every technological revolution is the spreadsheet. Autonomous transport will not be adopted on a mass scale until it is cheaper and more efficient than the work of a human driver. In the personal transport sector (robotaxis), the math is simple, though difficult to implement. The cost of a driver accounts for 50% to 70% of the price of a traditional taxi or Uber ride. By eliminating this factor, companies can drastically lower service prices for passengers while generating higher margins. However, to achieve this, the cost of purchasing and maintaining an autonomous fleet (including depreciation of expensive sensors, costs of servers processing data, and remote teleoperation support) must fall below the savings on wages. Currently, Waymo and other players are still subsidizing every kilometer driven, treating it as an investment in future market dominance.

The situation in trucking logistics is completely different. Here, the economic pressure is immediate. The transport industry around the world suffers from a structural shortage of drivers. This work is hard, requires multi-day separations from family, and is associated with rigorous regulations regarding working and rest times (a driver in the EU can drive a vehicle for a maximum of 9 hours a day). An autonomous truck does not have these limitations. It can drive for 20 hours a day, stopping only for refueling. Additionally, algorithms can drive the vehicle in an extremely smooth manner, optimizing fuel consumption by 10-15% and reducing wear on tires and brakes. For large logistics companies, the implementation of autonomous tractor units on inter-hub routes is a promise of huge savings that could translate into lower product prices on store shelves.

Possibilities and scenarios for the future

If, however, we manage to overcome these obstacles, the benefits of autonomous transport will be unprecedented. First and foremost, we are talking about a drastic increase in road safety. Over 90% of all road accidents are caused by human error – fatigue, alcohol, distraction by a phone, or recklessness. Machines do not drink, do not sleep, do not text while driving, and have a reaction time measured in milliseconds. Autonomous transport is a chance to save thousands of human lives every year.

Another aspect is the revolution in urban planning and social mobility. The elderly, the visually impaired, or people with disabilities will gain full transport independence. The model of vehicle ownership will also change. In cities, fleets of autonomous robotaxis available on demand will dominate. Since a car will be able to come for us in a few minutes, owning your own car, which sits uselessly in a parking lot for 95% of the time, will lose its meaning. This, in turn, will allow for the release of huge urban spaces currently occupied by parking lots and their transformation into parks or recreational zones.

In the area of logistics, autonomous trucks moving in optimized columns (platooning) will allow for lower fuel consumption, carbon dioxide emissions, and operating costs, operating 24/7 without the need for sleep breaks.

Summary: Realism instead of utopia

The development of autonomous cars teaches us humility in the face of the complexity of the world around us. It turned out that driving a vehicle in a chaotic, human environment is one of the most difficult tasks that modern artificial intelligence has faced. The era in which each of us will be able to buy a Level 5 car in a showroom and send it alone for groceries is still a distant prospect – we are likely decades away from it. Nevertheless, this revolution is happening before our eyes in an evolutionary way. Through the gradual implementation of Level 3 systems in premium cars, the development of autonomous transport corridors for trucks, and the slow but systematic expansion of robotaxi zones, we are entering a new era of mobility. It will not be a sudden jump, but a slow transformation that will ultimately change the face of our cities and our daily lives.

Sources

• https://www.sae.org/standards/content/j3016_202104/
• https://waymo.com/
• https://www.gmcruise.com/
• https://aurora.tech/
• https://ec.europa.eu/transport/themes/intelligent-transport-systems/its-general-presentation/automated-mobility_en
• https://www.nhtsa.gov/technology-innovation/automated-vehicles-safety
• https://www.mckinsey.com/industries/automotive-and-assembly/our-insights/the-future-of-mobility-is-autonomous-and-electric
• https://www.reuters.com/technology/autonomous-driving-technology-explained-2023-07-17/
• https://www.autonews.com/regulation-policy/us-states-take-different-paths-autonomous-vehicle-regulation
• https://www.transportation.gov/policy-initiatives/automated-vehicles/automated-vehicles-policy
• https://www.forbes.com/sites/johnkoetsier/2024/01/09/waymo-is-now-in-4-us-cities-and-growing-fast-heres-the-latest-on-its-robotaxi-service/
• https://www.techcrunch.com/2023/12/14/cruise-gets-green-light-to-resume-driverless-testing-in-california/