Part 5/15:
Limitations of Imitation Learning
While IL can produce impressive results—sometimes even surpassing average human performance—it has fundamental limitations:
Coverage of scenarios: IL depends on the diversity of the curated datasets. Rare or dangerous edge cases (like unanticipated pedestrian actions or unusual road configurations) are hard to cover exhaustively.
Inability to generalize fully: Mimicking human drivers does not guarantee handling novel or complex situations optimally.