The world of autonomous vehicles just got a lot more interesting, thanks to a thorough test conducted by Mark Rober, a YouTuber, engineer, and inventor. Rober put theAutopilot system of a Tesla Model Y, which relies solely on cameras and AI, to the test, comparing its performance to a vehicle equipped with LiDAR technology. The experiment simulated six different scenarios, and the results are quite revealing.
Tesla has developed its own Tesla Vision system, which uses a combination of cameras and AI to power its Autopilot feature, eliminating the need for LiDAR or other sensors from external companies. Elon Musk has confidently stated that Tesla Vision can learn and adapt to any road conditions, just like a human driver, and will accelerate the development of autonomous driving capabilities. However, it’s essential to note that the current Autopilot system is still at level 2, which means it’s an advanced driver-assistance system, not a fully autonomous one.
Critics argue that the camera-only approach has its limitations, especially considering Musk’s ambitious goal of achieving level 5 autonomy, where vehicles can drive themselves in all environments. The concern is that cameras alone might not be enough, and that’s where LiDAR comes in – by creating a highly accurate 3D map of the surroundings. However, Musk has opted not to use LiDAR, and this test aims to explore the implications of that decision.
In the six simulated scenarios, the Tesla Model Y was able to stop successfully when faced with stationary and moving dummies, as well as when subjected to interference from light. However, when the environment became more challenging, with thick fog or heavy rain, the vehicle failed to stop. On the other hand, the vehicle equipped with LiDAR successfully stopped in all scenarios, thanks to its ability to detect objects more accurately than the camera-based Tesla Vision system.
The most striking part of the test came when Rober introduced a fake wall with a painted road on it, reminiscent of the classic Wile E. Coyote cartoons. The LiDAR-equipped vehicle immediately stopped, recognizing the wall as a solid object, while the Tesla, relying on its cameras, drove right into it, fooled by the fake road.
These results highlight the limitations of a camera-only approach and suggest that Tesla’s Vision system may need further refinement, particularly when it comes to object detection in various scenarios. The test also sparks a debate about the potential drawbacks of relying solely on visual data, as opposed to the more accurate, sensor-based approach of LiDAR technology. It’s worth noting, however, that the test might have been designed to emphasize the weaknesses of the Tesla system, rather than showcasing its full capabilities in every possible scenario.
For more information on this topic, you can visit Electrek.
