Tesla's Initial Testing of Model Y Vehicles Involves Human Control Input
In the heart of Texas, Tesla's autonomous Full Self-Driving (FSD) software is being put to the test in a new robotaxi service, primarily using Model Y vehicles equipped with the FSD Beta software. This groundbreaking service, currently operating within a specific geofenced area in South Austin, is making waves in the autonomous vehicle industry.
The system, which relies on vision-only perception, processes real-time visual data via neural networks for object detection and tracking, combined with GPS and inertial measurement data to maintain accurate lane localization. Unlike traditional autonomous vehicles, Tesla's robotaxis do not use LiDAR or radar sensors, relying solely on eight external cameras and Tesla's custom AI computer.
Each robotaxi is supervised by a trained Tesla safety operator in the front passenger seat, ready to intervene physically via controls if the AI encounters a scenario it cannot resolve. In addition to this in-vehicle human oversight, Tesla has implemented a remote supervisory control layer with monitors in a central operations hub who track the real-time performance of the fleet. These remote monitors watch live camera feeds, sensor data, and system diagnostics to detect anomalies or unsafe behaviors the in-car safety monitor might miss, providing guidance or intervening by communicating with the in-vehicle operator.
The robotaxi service demonstrates good handling of complex driving scenarios like speed bumps and navigating interactions with human drivers. However, it still exhibits challenges such as mistakes in interpreting traffic signals, including stopping unnecessarily at stale yellow lights or misunderstanding red arrows. The AI software is a robust version of Tesla’s FSD v13 with roughly three times the neural network parameter count, indicating an advanced but not yet perfect system.
Notably, the Austin robotaxi service operates mostly without traditional safety drivers behind the wheel, instead depending on this combination of in-vehicle safety specialists and remote human oversight to ensure safety during real-world operations. This layered human oversight approach is critical during this early deployment phase to maintain safety while collecting data for iterative improvements.
The service, still categorized as a "test" by the city, is available only from 6 AM to 12 AM and is limited to those invited by Tesla. The first intervention by a human overseer in a Tesla robotaxi occurred on the second day of testing. The intervention was necessary when a Model Y attempting to park collided with a UPS delivery truck also trying to park in the same space. In another instance, a passenger took control of the driverless Tesla Model Y when it touched a parked car while passing.
Local Tesla fans have counted eleven robotaxis spotted in Austin, with only 10-20 autonomous Model Ys being used in the testing phase. The exact number of Model Ys and invited testers participating in the Austin test is unknown.
This combined approach reflects Tesla's cautious yet ambitious strategy for scaling autonomous ride-hailing safely and effectively. As the service continues to evolve, we can expect to see further improvements in the AI's ability to interpret traffic signals and navigate complex driving scenarios.
[1] Tesla AI Day 2021: Full Self-Driving Capability Update [2] Tesla's Full Self-Driving Beta: A Deep Dive into the Latest Update [3] Tesla's Full Self-Driving Beta: What's New and What's Next? [4] Tesla's Full Self-Driving Beta: How It Works and What It Means for the Future of Autonomous Driving
In this innovative landscape, Tesla's autonomous FSD software, not only revolutionizing the autonomous vehicle industry but also the finance sector through potential profits, is being tested in the robotaxi service. The integration of technology in transportation, including the use of neural networks, AI computers, and real-time data processing, showcases a fusion with the rapidly advancing field of technology and finance. Additionally, the remote supervisory control layer, linking the central operations hub to the testing fleet, presents an opportunity for investment in the future of transportation and technology.
