The dust has settled after a controversial YouTube video from Mark Rober, the engineer with more followers than Taylor Swift! All told, I think of this video as a win for the lidar industry. A very technical product is made relatable and well positioned in the automotive context. Check it out below and comeback for my thoughts here.
The Luminar lidar equipped vehicle performed flawlessly - an exciting, but expected result. The sensor's data is a deterministic 3D model of the vehicle's surroundings that is uninhibited by lighting or color camouflage. Every collision relevant obstacle is detected and when used by appropriate software, can help a car avoid collisions even with very short notice.
The controversy emerges from how poorly the Tesla vehicle performs, and more specifically, that Mark didn't ensure the vehicle was always operating in it's optimal performance mode ("Full Self Driving" or FSD). Some interesting technical topics emerge from this controversy and the scrutiny of millions of eyes globally.
Looking at testing by many different groups, Tesla cars perform much better in collision avoidance when Autopilot is engaged than when in normal human driving mode. Better still when FSD is engaged. The real reason behind this is only known by experts within Tesla, but it seems natural to me that the respective systems have different thresholds of pain associated with triggering phantom safety maneuvers (false-positives). Normal human driving mode is scrutinized by international standards where minimizing false-positive breaking is important and the effectiveness bar (true-positives) is actually very low. The pain of collisions to the automaker when in Autopilot and even more so in FSD is considered more problematic and so a higher acceptable occurrence of false-positive maneuvers is taken.
The trade between phantom braking reduction and collision avoidance should improve with continued advances in camera technology (both hardware and software) but as illustrated my Mark Rober, lidar changes the game.
The Luminar lidar equipped vehicle performed flawlessly - an exciting, but expected result. The sensor's data is a deterministic 3D model of the vehicle's surroundings that is uninhibited by lighting or color camouflage. Every collision relevant obstacle is detected and when used by appropriate software, can help a car avoid collisions even with very short notice.
The controversy emerges from how poorly the Tesla vehicle performs, and more specifically, that Mark didn't ensure the vehicle was always operating in it's optimal performance mode ("Full Self Driving" or FSD). Some interesting technical topics emerge from this controversy and the scrutiny of millions of eyes globally.
- Why wasn't FSD used? Mark's team asserts that FSD was not available since the testing was not conducted on a real roadway and the system needs an accessible destination. This makes sense give my understanding of how it works, but is a shame given how many people dismiss the result because of it.
- Why did Autopilot turn off just before hitting the wall? No clear assertions here, just conspiracies on both sides. "Mark turns it off at the last minute to ensure collision" vs "Autopilot disengages itself at the last minute so Tesla doesn't have to report Autopilot collisions."
- Why does it matter what mode the car is in? After all we're talking about emergency collision avoidance scenarios! This is my favorite question, because it likely exposes the topic of false-positive acceptance levels.
Looking at testing by many different groups, Tesla cars perform much better in collision avoidance when Autopilot is engaged than when in normal human driving mode. Better still when FSD is engaged. The real reason behind this is only known by experts within Tesla, but it seems natural to me that the respective systems have different thresholds of pain associated with triggering phantom safety maneuvers (false-positives). Normal human driving mode is scrutinized by international standards where minimizing false-positive breaking is important and the effectiveness bar (true-positives) is actually very low. The pain of collisions to the automaker when in Autopilot and even more so in FSD is considered more problematic and so a higher acceptable occurrence of false-positive maneuvers is taken.
The trade between phantom braking reduction and collision avoidance should improve with continued advances in camera technology (both hardware and software) but as illustrated my Mark Rober, lidar changes the game.
RSS Feed
