A two-stage trained transformer module, ADReFT, repairs unsafe autonomous driving decisions at runtime, fixing more collisions with smaller interventions than rule-based and anomaly baselines.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
ADReFT: Adaptive Decision Repair for Safe Autonomous Driving via Reinforcement Fine-Tuning
A two-stage trained transformer module, ADReFT, repairs unsafe autonomous driving decisions at runtime, fixing more collisions with smaller interventions than rule-based and anomaly baselines.