Beep-Beep! Watch Tesla's Autopilot Crash Into A Painted Wall In Wile E Coyote Test
If you need more evidence that camera-based image processing isn't at all a good idea for fully autonomous driving, go no further than this fun exposé comparing self-driving tech between a Tesla versus a LiDAR-equipped system in a Lexus. The multi-stage experiment culminates with both cars being driven towards a wall with a painted road to see which system successfully detects the anomaly. In short, the camera-based setup fails in a hilarious fashion, Wile E. Coyote-style.
Tech YouTuber (and former NASA engineer) Mark Rober recently dropped a video demonstrating what it takes to "fool a self driving car" that's equipped with camera-vision versus one that's equipped with LiDAR. Rober talked about the strengths and limitation of each system, but leaned heavily in support for LiDAR (Light Detection And Ranging) for its superior ability to map, process, and react in some of the most adverse situations.
In the child mannequin test, a Tesla Model Y on Autopilot braked when the mannequin was stationary, moving, and when the camera system was blinded by lights. However, the system failed in fog and heavy rain. In comparison, a LiDAR system mounted on a Lexus SUV successfully stopped in all those conditions.
The most dramatic test came from a wall painted with a fake road (and scenery) installed across an actual road. This test was made to illustrate how perception-based systems (like those used in Teslas) can't tell the difference between a fake road versus a real one, proven by how the Model Y drove straight through the wall without even blinking. On the other hand, the LiDAR-equipped Lexus wasn't fooled by the painting, instead properly identifying the wall for what it was.
As the video gained traction however, naysayers have been quick to point out that Rober's video used a LiDAR-equipped car provided by Luminar, a major LiDAR company. While it's hard to determine if there was any conflict of interest during the making of this video, the fact is that LiDAR systems ARE safer and more capable, especially in self-driving applications. However, LiDAR is more expensive to implement, which likely explains why Tesla has continued to push for camera-based systems in its vehicles. The company has promised Level 5 autonomy for the longest time, but that has continuously been delayed largely due to the camera system's difficulty in meeting regulatory/safety standards.
