Self-driving cars will use convolutional neural networks.

Judge at evaluation time; there is no stated horizon. True: self-driving or autonomous vehicle systems from major developers (for example Tesla, Waymo, Mobileye, NVIDIA DRIVE, Cruise) are documented as using convolutional neural networks in their perception or driving stacks, including as backbones within larger architectures. Mostly true: CNN use is documented, but only in some leading systems, or CNNs were later partly replaced by other architectures after being used. Mixed: CNN use appears only in research prototypes, not in deployed or major development vehicles. Mostly false: CNNs played only a marginal or experimental role. False: documentation shows self-driving systems do not use convolutional networks.

Set before the research began.

1. E1 Scalability in Perception for Autonomous Driving: Waymo Open Dataset ar5iv (arXiv paper by Waymo authors)

   “We use the same convolutional backbone architecture as the original paper REF:pointpillars_cvpr2018, with the slight exception that our Vehicle model matches our Pedestrian model in having a stride of 1 for the first convolutional block.”

2. E2 Tesla Full Self-Driving Technical Deep Dive iangreer.com · 15 December 2023

   “In Tesla's architecture, the backbone is RegNet + ResNet, the neck is BiFPN, and the head is HydraNet.”

3. E3 The Evolution of EyeQ Mobileye · 28 April 2026

   “which are optimized for a wide variety of computer-vision, signal-processing, and machine-learning tasks, including deep neural networks.”

4. E4 Edge AI in Cars: Why Tesla's Dojo Strategy… | Think Different Think Different (blog) · 21 October 2026

   “The neural networks are used for classification and object recognition, not for end-to-end decision making.”

The only explicit reference to convolutional networks is a backbone used in Waymo-authored models built for a research dataset [E1], which the report itself does not treat as a production stack. The other sources name Tesla's backbone architectures [E2] or mention neural networks in general [E3, E4] without saying they are convolutional. So the evidence does not establish CNN use in deployed or major development vehicles under the criteria.