Abstract

Trajectory prediction plays a pivotal role in the field of intelligentvehicles. It currently suffers from several challenges,e.g., accumulative errorin rollout process and weak adaptability in various scenarios. This paperproposes a parametric-learning recursive least squares (RLS) estimation basedon deep neural network for trajectory prediction. We design a flexible plug-inmodule which can be readily implanted into rollout approaches. Goal points areproposed to capture the long-term prediction stability from the globalperspective. We carried experiments out on the NGSIM dataset. The promisingresults indicate that our method could improve rollout trajectory predictionmethods effectively.

 

Quick Read (beta)