Identify the design factors responsible for cross-town transfer

Identify which components of the CORAL design— including the traffic-rule signals, stage-aware reward, curriculum schedules, LiDAR observation, and state representation—produce zero-shot transfer across CARLA towns.

Background

CORAL is trained in CARLA Town05 and evaluated in seven unseen towns, where it achieves variable success rates. However, the paper's ablations are conducted only within Town05, and the baseline methods are not evaluated across towns. Consequently, the reported transfer cannot be attributed to any particular design component.

References

Which part of the design produces that transfer is not resolved here: the ablations of Sections~\ref{subsec:exp_ablation} and~\ref{subsec:exp_lidar} vary one factor at a time but do so entirely within Town05, and the baselines were never run cross-town.

— CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving  (2608.14332 - Saleem et al., 14 Aug 2026) in Section 'Generalization Performance'

We therefore do not claim that the histogram contributes to the results reported in this paper, and the reason is the one that also makes the bin count irrelevant: the environment is static and the reference route runs along clear roadway, so an episode can be completed without ever localizing an obstacle. Establishing what the histogram is worth requires a benchmark on which an episode cannot be completed without it.

— CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving  (2608.14332 - Saleem et al., 14 Aug 2026) in Section 4.4, 'Ablation: The LiDAR Observation'