Commit Graph

1 Commits

Author SHA1 Message Date
Paul Huliganga ee91b8f9a3 feat(exp28): fine-tune exp26 best_model on generated-track with variable throttle
Warm-starts from exp26/best_model (best road model) and fine-tunes on
donkey-generated-track-v0 (shadows, trees) at LR=0.00005. Adds N_THROTTLE=3
variable throttle to force learning corner braking. 50k steps, eval on
mini-monaco (zero-shot) at completion.

Goal: visual diversity + throttle variation → better mini-monaco generalization.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-14 15:32:37 -04:00