Perceive, Predict, and Plan: Safe Motion Planning Through Interpretable Semantic Representations

TL;DR

This paper introduces an end-to-end neural network for autonomous vehicle motion planning that integrates perception, prediction, and planning with interpretable semantic representations, leading to safer and more human-like trajectories.

Contribution

It presents a novel differentiable semantic occupancy representation that aligns perception, prediction, and planning costs, enabling interpretable and safer motion planning.

Findings

Outperforms state-of-the-art planners in imitation accuracy.

Produces significantly safer trajectories in simulations.

Demonstrates effectiveness on large-scale driving data.

Abstract

In this paper we propose a novel end-to-end learnable network that performs joint perception, prediction and motion planning for self-driving vehicles and produces interpretable intermediate representations. Unlike existing neural motion planners, our motion planning costs are consistent with our perception and prediction estimates. This is achieved by a novel differentiable semantic occupancy representation that is explicitly used as cost by the motion planning process. Our network is learned end-to-end from human demonstrations. The experiments in a large-scale manual-driving dataset and closed-loop simulation show that the proposed model significantly outperforms state-of-the-art planners in imitating the human behaviors while producing much safer trajectories.