Neurodrive Puts a Fly Behind the Wheel

Field notes

The study maps lateral position, heading error, road curvature, and speed to nine steering-and-throttle choices, then updates a tabular Q-learning controller after each run. Its protocol fixes model steps at 0.12 seconds and ends an attempt after a lap, track exit, or 216 simulated seconds. Evaluation freezes learning and uses 20 held-out starts per controller, while the fly anatomy display mirrors the vehicle’s wheel and pedal commands.