Search

The collective perception service, which is in progress of standardization by the European Telecommunication Standards Institute, allows to share perception information among connected vehicles and road side units and thus can increase both safety and traffic efficiency. However, based on our practical experience from our research on infrastructure support of automated vehicles on a pilot installation in real traffic, in this work, we outline some drawbacks of the existing draft when applied to real-world environments. We observe that the strict cartesian representation does not fit well with typical models used to predict the motion of vehicles in automated driving applications. In theses cases, transformations and approximations are required, which increases the uncertainty about the perceived objects. In this work, we demonstrate the effect of such transformation errors using examples and propose an extension of the standard to prevent unnecessary transformations and approximations. Additionally, we show that the collective perception service can further be enhanced by allowing the optional transmission of motion predictions of perceived objects. That is, receivers benefit from saving computation time for object predictions and from the reception of high-quality motion predictions from road side units that are more accurate due to their knowledge of local peculiarities., acceptedVersion

Automated vehicles need to not only perceive their environment, but also predict the possible future behavior of all detected traffic participants in order to safely navigate in complex scenarios and avoid critical situations, ranging from merging on highways to crossing urban intersections. Due to the availability of datasets with large numbers of recorded trajectories of traffic participants, deep learning based approaches can be used to model the behavior of road users. This paper proposes a convolutional network that operates on rasterized actor-centric images which encode the static and dynamic actor-environment. We predict multiple possible future trajectories for each traffic actor, which include position, velocity, acceleration, orientation, yaw rate and position uncertainty estimates. To make better use of the past movement of the actor, we propose to employ temporal convolutional networks (TCNs) and rely on uncertainties estimated from the previous object tracking stage. We evaluate our approach on the public "Argoverse Motion Forecasting" dataset, on which it won the first prize at the Argoverse Motion Forecasting Challenge, as presented on the NeurIPS 2019 workshop on "Machine Learning for Autonomous Driving"., acceptedVersion