eduzhai > Applied Sciences > Engineering >

MultiXNet Multiclass Multistage Multimodal Motion Prediction

  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: One of the critical pieces of the self-driving puzzle is understanding thesurroundings of the self-driving vehicle (SDV) and predicting how thesesurroundings will change in the near future. To address this task we proposeMultiXNet, an end-to-end approach for detection and motion prediction baseddirectly on lidar sensor data. This approach builds on prior work by handlingmultiple classes of traffic actors, adding a jointly trained second-stagetrajectory refinement step, and producing a multimodal probability distributionover future actor motion that includes both multiple discrete traffic behaviorsand calibrated continuous uncertainties. The method was evaluated on alarge-scale, real-world data set collected by a fleet of SDVs in severalcities, with the results indicating that it outperforms existingstate-of-the-art approaches.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×