Trajectory Prediction in Autonomous Driving With a Lane Heading Auxiliary Loss

Abstract

This study investigates the use of trajectory and dynamic state information for efficient data curation in autonomous driving machine learning tasks. We propose methods for clustering trajectory-states and sampling strategies in an active learning framework, aiming to reduce annotation and data costs while maintaining model performance. Our approach leverages trajectory information to guide data selection, promoting diversity in the training data. We demonstrate the effectiveness of our methods on the trajectory prediction task using the nuScenes dataset, showing consistent performance gains over random sampling across different data pool sizes, and even reaching subbaseline displacement errors at just 50% of the data cost. Our results suggest that sampling typical data initially helps overcome the "cold start problem," while introducing novelty becomes more beneficial as the training pool size increases. By integrating trajectory-state-informed active learning, we demonstrate that more efficient and robust autonomous driving systems are possible and practical using low-cost data curation strategies.

Towards Safe and Sustainable Autonomous Vehicles Using Environmentally-Friendly Criticality Metrics

“…Denoting the angle of the nearest lane in time step i by θ * i , the yaw loss is the accumulated difference between θ * i and θ i . Note that [37] (Equation ( 6)) implies that the difference is non-zero, which contradicts their definition in [37] (Equation ( 3)). We suggest to use:…”

“…the vehicle is powered by electricity from the grid (37) where d is the distance travelled by the vehicle in the drive in kilometers.…”

“…Note that by the duality of reward terms and corresponding criticality metrics, one can also interpret reward terms like the Yaw loss [37], which penalizes non-optimal headings or the Off-road loss [38], which penalizes if the agent drives in the non-drivable area as criticality metrics.…”

“…as off-road loss for the whole trajectory. Note that the work in [37] also considers the yaw loss for intersections and for lane change where a pre-defined interval of heading differences is allowed so that the yaw loss is zero if the heading during lane change is contained in this interval.…”

Knowledge Augmented Machine Learning with Applications in Autonomous Driving: A Survey

“…Knowledge can also be integrated into tracking and trajectory prediction. In [261], the Yaw loss, an auxiliary differentiable heading loss that penalized angle differences between the optimal and the predicted headings, is proposed, where the case of road intersections is also respected. The authors in [518] propose an off-road loss for improving the movement prediction of traffic participants.…”