Control-Aware Prediction Objectives for Autonomous Driving

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.

Authors

• Rowan McAllister
• Blake Wulfe
• Jean Mercat
• Logan Ellis
• Sergey Levine
• Adrien Gaidon

Keywords

• Measurement
• Training
• System performance
• Predictive models
• Software
• Trajectory
• Behavioral sciences
• Autonomous Vehicles
• Downstream Effects
• Attention Model
• Importance Weights
• Trajectory Prediction
• Prediction Model
• Collision
• Prediction Error
• Probabilistic Model
• Optimal Control
• Attention Mechanism
• Kullback-Leibler
• Output Control
• Error Control
• Future Trajectories
• Attention Weights
• Prediction Loss
• Traditional Metrics
• Negative Log-likelihood
• Perceptual Errors
• Control Of Autonomous Vehicles
• Prediction Metrics
• Trajectories Of Agents
• Model-based Reinforcement Learning
• True Trajectory
• Map Information
• Prediction Probability
• Baseline Methods