Density-based Stochastic Reachability Computation for Occupancy Prediction in Automated Driving

TL;DR

This paper introduces a novel stochastic reachability framework for predicting occupancy in automated driving, directly solving the transport PDE to compute collision probabilities efficiently and accurately in real-time.

Contribution

It presents a gridless, nonparametric method based on characteristic curves for online occupancy prediction, improving computational efficiency over traditional grid-based approaches.

Findings

Effective online computation of collision probabilities

Demonstrated accuracy through numerical simulations

Applicable to real-time automated driving scenarios

Abstract

We propose a stochastic reachability computation framework for occupancy prediction in automated driving by directly solving the underlying transport partial differential equation governing the advection of the closed-loop joint density functions. The resulting nonparametric gridless computation is based on integration along the characteristic curves, and allows online computation of the time-varying collision probabilities. Numerical simulations highlight the scope the proposed method.