← back to legend
Physical / electrical concepts

Learned (RL) router

Reinforcement learning
A model that learns to route by scoring millions of its own attempts (Quilter, DeepPCB).glossary, Physical / electrical concepts
A model that learns to route by scoring millions of its own attempts (Quilter, DeepPCB). Option D.glossary, Ants: routing and our own terms (from Nov 2026)

How it looks

Drawing: an idealised patch of board
attempt 1: long detour, low score
A learning router starts out bad and is scored on every try.

How it works

THE SAME PROBLEM, TRIED OVER AND OVERtry 1: long way round. Score: low.try 5,000: shorter. Score: better.try 2,000,000: short and clean. Score: high.the model now proposes routes like this first timestarttarget

Where it fits

Option D of the three dials. What "scored better" means is the cost function. Compare hybrid routing, which needs no training.

← back to legend