End-to-end driving

A self-driving design in which one neural network maps raw sensor input directly to driving outputs like steering and speed.

• Classic systems split perception, prediction and planning into separate modules joined by hand-written rules.
• End-to-end systems learn the whole mapping from millions of clips of human driving, capturing subtle behaviour that rules miss.
• Performance scales with data and computing, favouring companies whose fleets collect driving video at scale, like Tesla.
• Failures are hard to diagnose or patch and safety is harder to prove, so many firms add rule-based checks around the network.