This article presents a hybrid-prediction integrated planning (HPP) system for autonomous driving. The HPP system addresses three major challenges in the integration of prediction and planning modules: inherent trade-offs by sole prediction, consistency between prediction patterns, and social coherence in prediction and planning. The HPP system includes three modules: a marginal-conditioned occupancy prediction to align joint occupancy with agent-wise perceptions; a game-theoretic motion predictor to model the interactive future among individual agents with their joint predictive awareness; and hybrid prediction patterns integrated with Ego Planner and optimized by prediction guidance. The HPP system demonstrated superior accuracy and consistency on the nuScenes dataset and the Waymo Open Motion Dataset and CARLA benchmark.
Publication date: 6 Feb 2024
Project Page: https://georgeliu233.github.io/HPP
Paper: https://arxiv.org/pdf/2402.02426
