Control-Aware Prediction Objectives for Autonomous Driving
Control-Aware Prediction Objectives for Autonomous Driving
复制标题
DOI：
10.48550/arxiv.2204.13319
复制
发表时间：
2022-04
期刊：
影响因子：
--
通讯作者：
R. McAllister;Blake Wulfe;Jean-Pierre Mercat;Logan Ellis;S. Levine;Adrien Gaidon
中科院分区：
文献类型：
--
作者：
R. McAllister;Blake Wulfe;Jean-Pierre Mercat;Logan Ellis;S. Levine;Adrien Gaidon
自动驾驶汽车软件通常被构建为单个组件（例如，感知、预测和规划）的模块化管道，以帮助将关注点分离为可解释的子任务。即使端到端训练是可能的，每个模块也有自己的一组目标，用于安全保证、样本效率、正则化或可解释性。然而，中间目标并不总是与整体系统性能一致。例如，优化轨迹预测模块的可能性可能更多地关注易于预测的代理，而不是安全关键或罕见的行为（例如，乱穿马路）。在本文中，我们提出了控制感知预测目标（capo），以评估预测对控制的下游影响，而不要求规划器是可微的。我们提出了两种类型的重要性权重来衡量预测可能性：一种使用代理之间的注意力模型，另一种基于在将预测轨迹交换为真实轨迹时的控制变化。通过实验，我们证明了我们的目标在郊区驾驶场景下提高了系统的整体性能。
Autonomous vehicle software is typically structured as a modular pipeline of individual components (e.g., perception, prediction, and planning) to help separate concerns into interpretable sub-tasks. Even when end-to-end training is possible, each module has its own set of objectives used for safety assurance, sample efficiency, regularization, or interpretability. However, intermediate objectives do not always align with overall system performance. For example, optimizing the likelihood of a trajectory prediction module might focus more on easy-to-predict agents than safety-critical or rare behaviors (e.g., jaywalking). In this paper, we present control-aware prediction objectives (CAPOs), to evaluate the down-stream effect of predictions on control without requiring the planner be differentiable. We propose two types of importance weights that weight the predictive likelihood: one using an attention model between agents, and another based on control variation when exchanging predicted trajectories for ground truth trajectories. Experimentally, we show our objectives improve overall system performance in suburban driving scenarios using the CARLA simulator.