Socially Compliant Control of Autonomous Vehicles with Application to Eco-Driving

TL;DR

The paper tackles the problem of AV control in mixed traffic by embedding a social psychology metric, Social Value Orientation (SVO), into the control design to account for how following HVs react to the AV's actions. It formulates a general optimization framework where the AV maximizes a weighted combination of its own payoff and the follower's payoff, with weights determined by the AV’s SVO angle $\varphi_{AV}$, and solves the resulting optimal control problem using Pontryagin's minimum principle. A concrete eco-driving example demonstrates how different SVO settings influence energy expenditure and follower vehicle speeds, revealing that more altruistic or prosocial AVs can improve traffic flow and follower responsiveness at the cost of higher AV energy use, particularly near intersections. The methodology is grounded, validated with real-world data from Highway 55, and offers a path toward socially aware AV behavior that can enhance safety and efficiency in mixed-traffic scenarios, while also outlining extensions to broader objectives and connectivity-enabled frameworks.

Abstract

Control design of autonomous vehicles (AVs) has mostly focused on achieving a prespecified goal for an individually controlled AV or for a swarm of cooperatively controlled AVs. However, the impact of autonomous driving on human-driven vehicles (HVs) has been largely ignored in AV controller synthesis, which could result in egoistic AV behavior detrimental to the safety of passengers and surrounding traffic. In this study we develop a general framework for socially compliant control design of AVs with a useful metric of social psychology, called social value orientation (SVO), allowing AVs to leverage their impact on the behavior of the following HVs. This is critical since AVs that behave in a socially compliant manner enable human drivers to comprehend their actions and respond appropriately. Within the proposed framework, we define the utilities of the controlled AV and its following vehicle, to be maximized in a weighted fashion determined by the AV's SVO. The utility maximization covers an array of design objectives given the goal of the AV and the benefits for the following HV stemming from the courtesy of socially compliant AV controls. An optimal control problem is then formulated to maximize the utility function defined, which is numerically solved using Pontryagin's minimum principle with optimality guarantees. The methodology developed is applied to synthesize socially compliant control for eco-driving of AVs. A set of numerical results are presented to show the mechanism and effectiveness of the proposed approach using real-world experimental data collected on Highway 55 in Minnesota.

Figures (7)

• Figure 1: Graphic illustration of human-AV interactions in mixed traffic consisting of HVs and AVs, where $i \in \mathcal{A}$ and $(i+1) \in \mathcal{H}$
• Figure 2: Graphic illustration of social value orientation (SVO) in angular notation, where $\varphi \approx \pi/2$, $\pi/4$, and $0$ correspond to 'altruistic', 'prosocial', and 'egoistic', respectively.
• Figure 3: Graphic illustration of a string of five vehicles, with the lead human-driven vehicle (HV) executing a real-world speed profile acquired on Highway 55 in Minnesota sun2022energy. The lead vehicle slows down at two consecutive intersections due to red signals.
• Figure 4: Speed profile of all vehicles considering different social preferences of AVs
• Figure 5: Trajectory profile of all vehicles considering different social preferences of AVs

Theorems & Definitions (3)

• Remark 3.1
• Remark 3.2
• Remark 4.1