Tesla is making progress on its full self-driving (FSD) software by testing it extensively in real-world conditions and adopting end-to-end neural networks by March 5, 2026. The software, now called full self-driving (supervised), had collected billions of miles of driving data from more than six million vehicles to boost security and performance.
Key Developments Within Real-World Testing
- End-to-end neural networks, a type of artificial intelligence model that processes input data and generates outputs without separate, hand-coded rules, are central to version 12. Tesla replaced more than 300,000 lines of C code with neural networks trained on videos of real drivers. This change allows the car to handle complex situations by learning from real-world examples instead of following preset instructions.
- Elon Musk claims FSD version fourteen should surpass human driving by 2 or 3 times, while version 15 aims for 10 times better performance.
- Global expansion. Tesla has started piloting FSD internationally. Recent trials in China and Australia show FSD surpassed local competitors on highways and in urban areas, despite lacking initial local training data.
- Shadow mode. Every Tesla operates FSD in shadow mode, where the software silently predicts actions and compares them to the driver’s choices; discrepancies are identified for further learning.
Technological Innovations
- A vision-only approach, unlike competitors who use LiDAR or RADAR. Tesla relies exclusively on eight cameras and artificial intelligence to interpret its environment.
- Tesla’s Dojo supercomputer processes vast video data, enabling fast updates and retraining of AI models.
- Occupancy networks are AI models that enable software to generate a real-time, three-dimensional bird’s-eye view of the road. This view maps the location of vehicles, pedestrians, and obstacles around the car, enabling the system to better anticipate the movement of surrounding objects and predict possible future events.
Continuing Difficulties & Scrutiny
- The NHTSA is still investigating FSD’s performance in low-visibility conditions and driver supervision.
- Despite its name, Full Self-Driving (FSD) is classified as a Society of Automotive Engineers (SAE) Level 2 system. SAE levels range from 0 (no automation) to 5 (full automation), and Level 2 means the vehicle can control steering and speed but still requires the driver to supervise and be ready to take over at any time.
- Edge case issues, such as phantom breaking and the interpretation of ambiguous line markings, remain significant engineering hurdles. Test continues to address these challenges in ongoing evaluations.
Tesla, the well-known electric car company known for its daring, sometimes debated automated driving technology, has just made a major change. to its full self-driving (FSD) software led by Elon Musk. The company is replacing its old F. FSD code with a new system as it prepares to launch its robotaxi service.
This change replaces the old code with a neural network-based system that interprets and adapts to real driving situations. It marks a significant departure from the previous rule-based model.
A Major Shift
Previously, the old FSD software relied on a complicated set of rules to help cars manage different driving situations. While effective at times, it often struggled with the unusual or unexpected events that drivers encounter on the road. In contrast, Tesla’s new FSD uses deep learning, letting the car learn from vast amounts of real-world driving data by observing it. In many different situations, the system can better understand its surroundings and adjust its driving.
With this transition to neural networks, several potential benefits emerge. The system can learn and improve over time without manual updates, continuously refining its decision-making as it experiences more types of driving situations.
However, moving to a neural network also entails new challenges. It’s important to ensure people understand how the system makes decisions, which helps build trust in the technology. Regulators, safety experts, and consumers all want to know how the software decides what to do, especially if something goes wrong.
People online have mixed opinions about the new system. CEO Michael Dell praised it on X, saying, “Super impressive. Tesla FSD v12.3 is like a human driver.” But some Reddit users had a different take: “I had my first experience with it, and it felt as if having a 15-year-old driver. It’s a very cool technical demo, but I feel like they have a long way to go to achieve their initial goal like at least 5 more years until they can even argue it’s level 4 or 5.”
Tesla’s History With Safety and Autonomy
Tesla began developing autonomous driving in 2014 with the launch of its Autopilot system. Initially, Autopilot featured adaptive cruise control and lane-keeping assist. Since then, Tesla has consistently refined and expanded Autopilot, aiming to eventually achieve full self-driving capability.
Tesla’s push regarding autonomy has faced multiple challenges and controversies. The company has been criticized for marketing Autopilot and FSD features in ways that some say could mislead customers about what the systems actually do. There have also been several well-known accidents involving Teslas using Autopilot, raising concerns about the technology’s safety and reliability.
Following these incidents, the National Highway Traffic Safety Administration (NHTSA) initiated an investigation into Tesla’s Autopilot system. In a statement, the NHTSA emphasized the importance of ensuring that vehicles with automated driving systems operate safely and as intended, necessary to maintain public trust and confidence in these technologies.
Tesla’s decision to fully update its FSD software has drawn mixed reactions from specialists and safety advocates. Some people worry about the risks of rolling out new technology on a larger scale, especially given Tesla’s mixed track record with self‑driving technology.
Experts say renaming “full self‑driving” to “supervised full self‑driving” clarifies the system’s capabilities. Pratik Chaudhari, an engineering professor and former developer of autonomous taxis, noted that drivers have always been required to provide active supervision. The new name emphasizes this fact.
Obstacles in Achieving Full Autonomy
Chowdhury pointed out the shortcomings of current automotive driving technology, stating that there are still regular incidents in which Teslas and assistive autonomous cars from other car makers have behaved in an unsafe manner. The driver is expected to remain alert and intervene in the event of such incidents. He stressed the challenges of handling unforeseeable human behavior and of ensuring a car is 99.99% safe, given the vast diversity of situations that can occur on the roads.
A major challenge for total autonomy is getting driverless cars to handle unusual or unexpected situations. Machine learning and computer vision have improved significantly in recent years, but these technologies still have a long way to go before they can match the way humans adapt and make choices on the road.
Self-driving companies use various strategies to improve safety and reliability with real-world testing, exposing software to a wide range of situations. As Chaudhary noted, ensuring full safety remains difficult due to the sheer diversity of scenarios vehicles may encounter.
Alternative Approaches to Automated Driving Technology
Tesla mainly uses visible-light cameras and neural-network software for its automated driving technology, but other companies are adopting different approaches. Shawn Taikratoke, CEO of autonomous mobility startup Mozee, said he was impressed by Tesla’s choice to rebuild its FSD software. He explained that Tesla’s daring decision to rebuild its full self-driving suite reflects a culture that focuses on speed, originality, and responsiveness qualities that are critical as it works toward ambiguous, ambitious targets, such as deploying robotaxis.
Mozee takes a different approach by using a broader range of sensors, including radar and infrared sensors. This ensures their vehicles work well in many settings. Taikratoke said our diverse sensor approach ensures our vehicles can function reliably across a wide range of environments. These include the planned pathways of university campuses and the unforeseeable streets of metropolitan centers. By maintaining this flexible and thorough approach, we ensure that our technology is adaptable and scalable. This reflects our customers’ diverse demands and the environments in which they operate.
Mozee believes that for self-driving technology to succeed, vehicles and infrastructure must communicate seamlessly to form a connected network that enhances safety and efficiency. Taikratoke stressed the need for teamwork and partnerships to move the business forward. He said as we grow, this adaptivity will be critical not only to fulfilling but surpassing the expectations of our partners and the communities we serve. We are excited to partner with industry leaders like Tesla to shape the future of transportation and create a safer, more effective world for all.
As the industry evolves, companies like Tesla, Mozee, and Waymo are leading the way in developing flexible, expandable solutions that prioritize safety, efficiency, and real-world use, although some people are concerned about Tesla’s complete overhaul of its FSD software. It demonstrates the company’s commitment to innovation and its readiness to tackle the challenges of building fully self-driving vehicles.
To make fully self-driving vehicles a reality, more technological advances are needed, along with progress in regulations and community acceptance. Governments and regulators must set explicit rules and standards for the development, testing, and use of self-driving vehicles. Building community confidence is also important, since people need to feel confident in the safety and performance of these systems before they become common.
In a 2020 interview, Elon Musk expressed optimism about self-driving technology. He said, “I am extremely confident that we will have the basic functionality for level 5 autonomy completed this year.” Still, he admitted there are challenges. “There are many small problems that need to be solved, and there’s the challenge of solving all those small problems and putting the whole system together.”
As the self-driving industry grows, companies, researchers, and regulators will need to work together to develop self-driving technology safely and responsibly. By joining forces to address the technical, legal, and social challenges of self-driving vehicles, we can help establish a future in which self-driving transit changes how we live, work, and travel.
As Chowdhari noted, achieving safe driving is also a slow march of the nines. This means making the system 99%. 99.9% safe, then 99.999%, and so on. The aim is always to improve and solve rare problems. It requires technological progress as well as progress on policy and infrastructure.
Source: Tesla’s new self-driving software throws out its old code entirely
