Mark Rober, a former NASA engineer and renowned science communicator, has made a very entertaining video that addresses one of the most debated topics in autonomous driving: Tesla Autopilot’s camera-only approach vs. using lidar for autonomous vehicles.
Six Tests to Compare Tesla’s Camera-only Approach to Lidar
Rober designed six test: detecting a kid, rapid response, (i.e., a fake kid running out in front of the car), foggy conditions, rainy conditions, and the ultimate test of all – a fake wall reminiscent of a classic Wile E. Coyote gag. The Lexus SUV equipped with Luminar lidar passed all six tests. Tesla only passed three. I won’t spoil all the details, though.
Elon Musk, Famously Anti-Lidar
Elon Musk has long championed a vision of autonomous driving that mirrors human perception, advocating for a system that relies solely on cameras, akin to how we navigate the world. This philosophy led Tesla to eschew technologies like lidar and radar, focusing instead on a camera-based approach for their vehicles’ Autopilot and Full Self-Driving (FSD) systems. Musk has been quoted saying, “Humans drive without shooting lasers out of their eyes. Why should autonomous cars be any different?”
Tesla’s Vision-based Approach
However, Rober’s evaluations, while not being the most scientific, have highlighted potential shortcomings with Tesla’s Autopilot. Tesla’s Autopilot system is built around a suite of eight cameras strategically positioned to provide a 360-degree view of the vehicle’s surroundings. These cameras feed visual data into an onboard neural network, which processes the information to interpret the environment and make driving decisions. Tesla’s approach emphasizes a “pure vision” methodology, relying exclusively on video input without supplementary sensors like lidar or radar. This design choice is rooted in Musk’s belief that a camera-only system can achieve full autonomy by mimicking human visual perception.
Why Most Autonomous Vehicles Use Lidar
The majority of autonomous vehicle systems incorporate lidar as a core sensing technology. This video demonstrate a Lexus SUV with Luminar lidar. Lidar is an active sensing technology, emitting laser pulses and detecting their return to create precise, three-dimensional maps of the environment. Automotive lidar sensors are designed for both long-range and short-range detection capabilities, enhancing the vehicle’s ability to perceive obstacles and navigate complex scenarios. Commonly, lidar is combined with other sensors, providing a multi-faceted view of the surroundings, potentially offering a more robust solution for autonomous driving.
Tesla Autopilot vs. Lidar Put to the Test
The tests used in this video are a bit more playful and entertaining than a proper scientific comparison – especially the Wile E. Coyote gag of driving through a fake wall – but I think the results are worthy of consideration. The lidar-equipped vehicle decisively outperforms Tesla’s Autopilot, proving its reliability across a range of challenging scenarios. While some of Rober’s test conditions may seem unconventional, they highlight a crucial lesson in autonomous driving: real-world conditions are often unpredictable and fall outside the norm. The ability to navigate these edge cases effectively isn’t just a bonus—it’s a key factor that will determine the long-term adoption and success of self-driving technology.
In the first half of the video, Rober provides a basic understand of lidar to those who are not familiar with the technology and showcases applications of lidar technology by scanning iconic attractions like Space Mountain and the Haunted Mansion. I found this section equally interesting, I so recommend watching the entire video – perhaps at 2x speed in some sections ;-).
