Phil Trubey(@PTrubey):This is an awesome video and everyone who is wondering about the tech behind driverless cars should watch it. The only thing I would add is that in real world production systems you always have constraints. For a car, this means you can’t use an expensive GPU cluster for the AI. You have a strict limit to the amount of compute you have. Meaning that if you have Lidar and radar, you are starving vision of compute because some of that compute now has to go to analyzing Lidar, radar and doing the sensor fusion. You’ve just made a tradeoff deciding that your vision system isn’t going to be as good as on a car with a vision only system.

2026.10.17 23:11

This is an awesome video and everyone who is wondering about the tech behind driverless cars should watch it. The only thing I would add is that in real world production systems you always have constraints. For a car, this means you can’t use an expensive GPU cluster for the AI. You have a strict limit to the amount of compute you have. Meaning that if you have Lidar and radar, you are starving vision of compute because some of that compute now has to go to analyzing Lidar, radar and doing the sensor fusion. You’ve just made a tradeoff deciding that your vision system isn’t going to be as good as on a car with a vision only system.

Genma_Jp@nymbusjp

2026.10.17 03:35

15 years building multi-sensor fusion taught me one thing: fusion is a crutch. 🔹 More sensors ≠ more information 🔹 LiDAR’s and radar's physics: rain, blooming, range-Doppler grid, and target separability 🔹 Why only vision sees meaning 🔹 Why fusion is a crutch See video 👇