Tesla Autopilot now allows the car to sense the area around it, thanks to the development of Occupancy Networks. Ashok Elluswamy, Tesla’s Autopilot Software Director, shared a detailed thread on Twitter. about one last workshop The autopilot team organized. He also shared the workshop on Twitter.
At CVPR this year, he presented some of the latest work from the Tesla Autopilot team, specifically “Occupancy Networks” – our approach to solving general obstacle detection and using it to achieve complex collision avoidance. The full talk is here: https://t.co/wpGXlNHaWl (1/12)
– Ashok Elluswamy (@aelluswamy) 21 August 2022
In the video and Twitter Thread, Ashok explained how Tesla literally developed Occupancy Networks to give the car a sense of its surroundings. Humans have the ability to understand objects around them at any given time. Is this car moving on the road at a slow speed or at a fast speed? As a pedestrian, do I have enough time to cross the street before crashing? What’s in the middle of the road? What is falling from the sky? I have to get out of the way.
These reactions to scenarios and snap decisions come naturally to people. Tesla’s Autopilot Team is working on programming vehicles to do the same, and it will save lives. Imagine being able to accurately sense the car’s surroundings while the driver isn’t even paying attention. An example is sudden undesired accelerations (SUA). Ashok noted that Autopilot prevents about 40 of such accidents every day.
The workshop was held in June this year. Computer Vision and Pattern Recognition Conference (CVPR.) in New Orleans. Ashok explained that the team has developed Occupancy Nets that allow it to predict the volumetric occupancy of everything around the car.
Ashok explained that typical approaches such as image-space segmentation of empty space or depth by pixel have many problems. The solution to these problems is Occupancy Networks.
In other words, Occupancy Networks allows the vehicle to detect the space around it and determine if it can move in that space. For example, if a UFO suddenly crashes in front of you while driving, you would react quickly in the safest way possible. This is what the Autopilot Team has trained the software to do.
Ashok shared details of how Occupancy Networks uses Neural Radiance Fields (NeRFs). “The occupancy representation of these networks allows for differentiable rendering of images (based on the Neural Brightness Fields study). However, unlike typical NeRFs per scene, these occupancy nets generalize across scenes.”
These predictions are already used to avoid many collisions. For example, Autopilot avoids ~40 collisions per day where human drivers accidentally step on the accelerator pedal 100% instead of the brakes. In the video, Autopilot automatically brakes to save this person’s legs (7/12) pic.twitter.com/XtMssPT9cM
– Ashok Elluswamy (@aelluswamy) 21 August 2022
You can read Ashok. The full Twitter thread is here and you can watch the presentation here. we are a a little over a month Before Tesla’s AI Day and I’m sure Tesla will share more about the life-saving technology it’s working on and the Optimus Bot.
Dr. lately know all published and video shared his thoughts on the new 10.69 update and Occupancy Network.
In a message on Twitter he told me, “The beauty of Occupancy Networks is that the car doesn’t have to know. What objects he sees, they just need to know has It’s there to avoid them!”
Note: Johnna is a Tesla shareholder and supports its mission.
Your feedback is important. If you have any comments, concerns, or see a typo, you can email me at: johnna@teslarati.com. You can also reach me on Twitter @JohnnaCrider1
