This paper addresses the challenge of high-quality labeled data requirements in end-to-end autonomous driving (AD) and introduces a planning-oriented active learning method called ActiveAD. The main bottleneck in end-to-end AD is the need for high-quality labeled data, such as 3D bounding boxes and semantic segmentation, which are expensive to manually annotate. The data often suffers from a long-tailed distribution, with a large portion being trivial and only a few cases being safety-critical. To tackle this issue, ActiveAD progressively annotates raw data based on diversity and usefulness criteria for planning routes. Empirical results show that ActiveAD outperforms general active learning methods and achieves comparable performance to state-of-the-art end-to-end AD methods using only 30% of the nuScenes dataset. The method introduces Ego-Diversity for initial sample selection, and three metrics—Displacement Error, Soft Collision, and Agent Uncertainty—for incremental selection. Extensive experiments validate the effectiveness of ActiveAD, demonstrating its ability to select informative samples and improve planning performance. The contributions include a novel approach to data-centric AD and a detailed analysis of the proposed metrics.This paper addresses the challenge of high-quality labeled data requirements in end-to-end autonomous driving (AD) and introduces a planning-oriented active learning method called ActiveAD. The main bottleneck in end-to-end AD is the need for high-quality labeled data, such as 3D bounding boxes and semantic segmentation, which are expensive to manually annotate. The data often suffers from a long-tailed distribution, with a large portion being trivial and only a few cases being safety-critical. To tackle this issue, ActiveAD progressively annotates raw data based on diversity and usefulness criteria for planning routes. Empirical results show that ActiveAD outperforms general active learning methods and achieves comparable performance to state-of-the-art end-to-end AD methods using only 30% of the nuScenes dataset. The method introduces Ego-Diversity for initial sample selection, and three metrics—Displacement Error, Soft Collision, and Agent Uncertainty—for incremental selection. Extensive experiments validate the effectiveness of ActiveAD, demonstrating its ability to select informative samples and improve planning performance. The contributions include a novel approach to data-centric AD and a detailed analysis of the proposed metrics.