Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces BucketKD, a novel knowledge distillation framework designed for end-to-end motion planning in autonomous driving, addressing the challenge of large model sizes in resource-constrained environments. By discretizing critical environmental variables into adaptive buckets and incorporating a safety-aware waypoint attention mechanism based on time-to-collision metrics, BucketKD enhances both the efficiency and safety of motion planners. Experimental results demonstrate that BucketKD outperforms existing methods in planning accuracy and safety while achieving significant model compression.
Compact motion planners can be both efficient and safety-aware, with BucketKD achieving superior performance compared to state-of-the-art methods.
End-to-end motion planning has emerged as a promising paradigm in autonomous driving, directly mapping raw sensor data to control commands via deep neural networks. Despite its advantages, its large model size hinders deployment in resource-constrained platforms. In this paper, we present BucketKD, a bucket-based knowledge distillation framework that yields compact and safety-aware end-to-end planners. Compared to the state-of-the-art approach, which relies on simplified planning state representations, BucketKD discretizes critical environmental variables into adaptive buckets that capture richer scene semantics while preserving efficiency. In addition, we design a safety-aware waypoint attention mechanism that evaluates each waypoint's risk level by accounting for both obstacle proximity and relative motion through a time-to-collision (TTC) formulation widely used in transportation research. This enables the student model to better retain safety-critical behaviors during distillation. Extensive experiments in CARLA using the Bench2Drive dataset show that BucketKD significantly outperforms the state-of-the-art in both planning accuracy and safety while maintaining strong compression ratios.