Tesla Autopilot received new visuals in the 2020.16.2 over-the-air software update, including a more realistic animation of moving pedestrians. Tesla owner and YouTube channel host Julian's Random Project tested out Autopilot's new capabilities by playing a four-hour video of a drive through the streets of Hawaii.
Julian's Tesla detected and accurately rendered pedestrians that were "walking" in front of the vehicle. The Tesla owner explained that before 2020.16.2, pedestrians simply moved across the screen of his car's infotainment system with less fluidity. The OTA update more accurately depicts pedestrians walking in front of Tesla vehicles, capturing the gait of the person and smoothly animating people as they walk from one side of the street to the other. Even people on bicycles are animated better in the 2020.16.2 update.
Julian's video also showed how Tesla's vision-based approach worked, revealing how Autopilot reacted and continuously adapted to the environment it "saw" playing out in the video it was watching. The Tesla owner played a video of someone taking a drive around Hawaii, and Autopilot reacted to the road it saw on the screen, even though the actual vehicle was physically parked inside his garage.
Julian's test may have confused Autopilot a little bit, but it did reveal how well Tesla's driver-assist software could easily adapt to new roads and paths it had never driven on before. Andrej Karpathy talked about Tesla's vision-based approach for autonomous driving in a presentation at Matroid Scaled Machine Learning Conference 2020 in February.
In his talk, the AI Director discussed Tesla's ongoing improvements with Autopilot and the company's progress in its Full Self-Driving initiative. Karpathy particularly mentioned Tesla's work with 3D labeling and the AI team's efforts to figuring out how to get a birds-eye view of the car's environment for Autopilot to use.
ARK Invest forecasted that Tesla's vision-based approach, in conjunction with 3D labeling, would ultimately give the EV automaker a decisive advantage in the ride-hailing service market.
"In addition to vague local maps and its camera-based approach, 3D video labeling separates Tesla from its competitors, enabling the recognition of corner cases in solving for full autonomy. We believe Tesla's approach is highly differentiated and will be almost impossible for a competitor to replicate. While autonomous driving is an extremely complex problem to solve, Tesla could enjoy a near-monopoly in autonomous ride-hailing if it is successful," wrote ARK in its newsletter.
Featured Image Credit: JuliansRandomProject/YouTubeFollow @PurplePanda88