LEAD: Minimizing Learner-Expert Asymmetry in End-to-End Driving

2025-12-23

AI Summary

This work addresses systematic asymmetries between expert and student policies in end-to-end imitation learning for autonomous driving—specifically, perceptual visibility disparity, behavioral uncertainty mismatch, and ambiguous navigation intent. We quantitatively characterize, for the first time, the coupled impact of these three asymmetries on closed-loop driving performance. To mitigate them, we propose an Expert–Student Alignment Framework coupled with a multi-stage intent enhancement mechanism. Our approach integrates multimodal perception supervision, explicit route encoding, occlusion-robust training, and sim-to-real joint optimization, implemented atop the TransFuser v6 architecture to yield interpretable intent modeling, robust perception, and behaviorally consistent policy learning. On CARLA Bench2Drive, our method achieves 95 DS; on Longest6~v2 and Town13, it surpasses prior state-of-the-art by over 2×. Consistent improvements are also observed on NAVSIM and Waymo Vision-Based benchmarks.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Visual Reasoning & Symbolic Representations

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Simulators can generate virtually unlimited driving data, yet imitation learning policies in simulation still struggle to achieve robust closed-loop performance. Motivated by this gap, we empirically study how misalignment between privileged expert demonstrations and sensor-based student observations can limit the effectiveness of imitation learning. More precisely, experts have significantly higher visibility (e.g., ignoring occlusions) and far lower uncertainty (e.g., knowing other vehicles' actions), making them difficult to imitate reliably. Furthermore, navigational intent (i.e., the route to follow) is under-specified in student models at test time via only a single target point. We demonstrate that these asymmetries can measurably limit driving performance in CARLA and offer practical interventions to address them. After careful modifications to narrow the gaps between expert and student, our TransFuser v6 (TFv6) student policy achieves a new state of the art on all major publicly available CARLA closed-loop benchmarks, reaching 95 DS on Bench2Drive and more than doubling prior performances on Longest6~v2 and Town13. Additionally, by integrating perception supervision from our dataset into a shared sim-to-real pipeline, we show consistent gains on the NAVSIM and Waymo Vision-Based End-to-End driving benchmarks. Our code, data, and models are publicly available at https://github.com/autonomousvision/lead.

Problem

• Minimizes learner-expert asymmetry in driving simulators
• Addresses misalignment between expert demonstrations and student observations
• Improves closed-loop performance in end-to-end driving policies

Innovation

• Reduces expert-student visibility and uncertainty gaps
• Enhances navigational intent specification beyond single points
• Integrates perception supervision in sim-to-real pipeline