Tesla's Challenge in Achieving Fully Driverless and Driverless Taxis
TapTechNews June 19 - At Tesla's 2024 annual shareholders' meeting, CEO Elon Musk talked about a major problem plaguing Tesla's Full Self-Driving (FSD). This problem confirms the speculation of industry insiders two years ago and may affect Tesla's goal of achieving driverless taxis.
Musk said that as the FSD system continues to progress, it becomes increasingly difficult to determine which AI model is better. Because when there is only one instance of needing human intervention after driving thousands of miles how to quickly assess the performance of a new model becomes a problem.
These different AI models don't solve all problems perfectly, Musk admitted. One model solves problem A but may bring problem B.
The Cleantechnica editor Zachary Shahan推测（predicted） two years ago that there might be this seesaw problem in the Tesla FSD system, that is, the system may introduce new problems in the process of correcting existing problems, resulting in停滞不前（stagnation） of the overall effect.
How does Tesla solve this problem? We are solving this problem in many ways, including simulation tests and shadow mode operation, Musk said. Having some vehicles not enable the FSD function can actually help. We can compare the driving behavior of the new model with the user behavior through shadow mode to find out the advantages and disadvantages of different models.
Tesla can conduct tests through its millions of vehicles to compare the differences between the predicted behavior of the AI model and the actual driving behavior of the users to determine which model performs better.
The current biggest limitation is not the training data, but the efficiency of testing the AI model, Musk admitted. And how to cleverly judge whether a new model is better. For example, we know that some specific intersections are very complex... All models perform well on good roads without complex intersections. Therefore, we have selected thousands of complex intersections in the US specifically to test the performance of new models.
TapTechNews noted that Musk also mentioned that the full potential of the self-driving hardware Hardware 4 has not been fully exploited yet and plans to upgrade it later this year.
In conclusion, solving the seesaw problem and rapidly iterating the FSD model will be important challenges for Tesla to achieve fully driverless and driverless taxis.