Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Stackelberg Meta-Learning Based Shared Control for Assistive Driving

Yuhan Zhao, Quanyan Zhu

TL;DR

This paper addresses robust shared control for assistive driving under environmental uncertainty and bounded rationality in human drivers. It introduces a Stackelberg meta-learning framework that couples a dynamic Stackelberg game (ADAS as leader, human as follower) with a quantal response model to capture decision noise, and learns a generalized follower utility across driver types that can be rapidly adapted to a specific driver using few samples. In a three-lane obstacle avoidance scenario, the adapted driver model enables the ADAS to efficiently guide diverse drivers to the target while reducing travel time and showing resilience to bounded-rationality errors. The work provides a practical pathway for fast, driver-specific planning in shared control, with potential impact on safety and user experience.

Abstract

Shared control allows the human driver to collaborate with an assistive driving system while retaining the ability to make decisions and take control if necessary. However, human-vehicle teaming and planning are challenging due to environmental uncertainties, the human's bounded rationality, and the variability in human behaviors. An effective collaboration plan needs to learn and adapt to these uncertainties. To this end, we develop a Stackelberg meta-learning algorithm to create automated learning-based planning for shared control. The Stackelberg games are used to capture the leader-follower structure in the asymmetric interactions between the human driver and the assistive driving system. The meta-learning algorithm generates a common behavioral model, which is capable of fast adaptation using a small amount of driving data to assist optimal decision-making. We use a case study of an obstacle avoidance driving scenario to corroborate that the adapted human behavioral model can successfully assist the human driver in reaching the target destination. Besides, it saves driving time compared with a driver-only scheme and is also robust to drivers' bounded rationality and errors.

Stackelberg Meta-Learning Based Shared Control for Assistive Driving

TL;DR

This paper addresses robust shared control for assistive driving under environmental uncertainty and bounded rationality in human drivers. It introduces a Stackelberg meta-learning framework that couples a dynamic Stackelberg game (ADAS as leader, human as follower) with a quantal response model to capture decision noise, and learns a generalized follower utility across driver types that can be rapidly adapted to a specific driver using few samples. In a three-lane obstacle avoidance scenario, the adapted driver model enables the ADAS to efficiently guide diverse drivers to the target while reducing travel time and showing resilience to bounded-rationality errors. The work provides a practical pathway for fast, driver-specific planning in shared control, with potential impact on safety and user experience.

Abstract

Shared control allows the human driver to collaborate with an assistive driving system while retaining the ability to make decisions and take control if necessary. However, human-vehicle teaming and planning are challenging due to environmental uncertainties, the human's bounded rationality, and the variability in human behaviors. An effective collaboration plan needs to learn and adapt to these uncertainties. To this end, we develop a Stackelberg meta-learning algorithm to create automated learning-based planning for shared control. The Stackelberg games are used to capture the leader-follower structure in the asymmetric interactions between the human driver and the assistive driving system. The meta-learning algorithm generates a common behavioral model, which is capable of fast adaptation using a small amount of driving data to assist optimal decision-making. We use a case study of an obstacle avoidance driving scenario to corroborate that the adapted human behavioral model can successfully assist the human driver in reaching the target destination. Besides, it saves driving time compared with a driver-only scheme and is also robust to drivers' bounded rationality and errors.
Paper Structure (4 sections, 1 figure)

This paper contains 4 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Shared Control as Dynamic Stackelberg Games

Figures (1)

  • Figure 1: Illustration of Stackelberg shared control framework. The ADAS leverages meta-learning (1) to compute a meta utility model for different human drivers and adapts it to a driver-specific model to perform Stackelberg collaboration (2) for assistive driving tasks. We verify the algorithms using an obstacle avoidance driving scenario (3).

Theorems & Definitions (1)

  • remark thmcounterremark