Using Large Models to Oversee Smaller Models' Future Actions: Early On-Policy Distillation for Autonomous Driving

abursuc · x · 2026-09-16

In discussions around SSAD 2026, researcher abursuc outlined an approach for bringing LLM-style "think-ahead" to autonomous driving: a large teacher model supervises the future actions of a smaller student model. On closer inspection, this can be viewed as an early form of on-policy distillation applied to driving tasks.

Related event: SSAD 2026 Explores LLM-Style Think-Ahead Supervision for Autonomous Driving(3 posts)→

Original post →

More from Research

Research channel →