ICR-Drive
Instruction Counterfactual Robustness for End-to-End Language-Driven Autonomous Driving
ICR-Drive studies instruction counterfactual robustness in language-conditioned autonomous driving. The project evaluates how meaning-preserving, ambiguous, noisy, and misleading instruction variants affect closed-loop driving behavior in CARLA.
This work focuses on diagnostic evaluation for human-centered autonomous driving systems, with an emphasis on robustness, interpretability, and safety under linguistic variation.