AssistMe: Policy iteration for the longitudinal control of a non-holonomic vehicle

TL;DR

AssistMe introduces a model-based semi-autonomous control algorithm for safe vehicle navigation, utilizing policy iteration and driver intention modeling, demonstrating improved safety over baseline controls in tests with healthy participants.

Contribution

The paper presents a novel policy iteration approach for longitudinal vehicle control that incorporates stochastic driver models and requires limited online computation.

Findings

ASC outperforms baseline rule-based control with statistical significance

The algorithm effectively integrates vehicle dynamics and driver intentions

Limited online computation with pre-computed maps suffices for real-time control

Abstract

In this article we design a physically-inspired model-based assist-as-needed semi-autonomous control (ASC) algorithm to address the problem of safely driving a vehicle (a power wheelchair) in an environment with static obstacles. Once implemented online, the proposed algorithm requires limited computing power and relies on pre-computed (offline) maps (look-up tables). These are readily available by implementing policy iteration that minimizes the expected time to termination (safely stopping near an obstacle), by taking into account: (i) the vehicle dynamics; (ii) the drivers' intention modeled as three separate stochastic processes. We call them the expert driver, the naughty child and the blind driver models. A study with healthy participants confirmed that ASC outperforms a baseline rule-based control (a statistically significant result).