End-to-end (E2E) autonomous-driving planners trained by imitation are prone to statistical shortcuts: they associate scene elements that merely co-occur with expert actions (such as a roadside object or building facade) with driving decisions, instead of the variables that causally determine them. This causal confusion silently compromises reliability in long-tail scenarios and is difficult to detect, as prevailing open-loop metrics (L2 displacement and collision rate) are dominated by ego status and do not indicate whether a planner depends on spurious cues. Existing remedies based on causal-intervention training require retraining large models and cannot audit a planner that is already deployed. We present CADET, a training-free framework that audits, benchmarks, and repairs spurious reliance in pretrained E2E planners without any parameter update.
Blogger's Review: The introduction of the CADET framework offers an innovative approach to addressing causal confusion in end-to-end autonomous driving, avoiding the cumbersome process of model retraining. This method effectively audits and rectifies the shortcomings of deployed systems, providing significant practical value and research significance. Its training-free nature also facilitates easier application in real-world environments.