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

Shian Wang

2024-01-02

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.

Systems and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the design of social - compliant control for autonomous vehicles (AVs) in mixed traffic flow. Specifically, the paper focuses on how to make the behavior of autonomous vehicles not only achieve their own goals, but also take into account the impact on surrounding human - driven vehicles (HVs), thereby improving overall traffic safety and efficiency. Most previous studies have focused on the control of a single autonomous vehicle or a group of autonomous vehicles, but have ignored the impact of autonomous vehicle behavior on human - driven vehicles, which may lead to selfish autonomous vehicle behavior and endanger the safety of passengers and surrounding traffic. For this reason, this paper proposes a general framework based on social value orientation (SVO), allowing autonomous vehicles to utilize their influence on the behavior of subsequent human - driven vehicles to achieve social - compliant control design.

### Main contributions of the paper
1. **Proposing a general framework based on social value orientation (SVO)**: This framework allows the inclusion of multiple design goals according to actual needs, enabling autonomous vehicles to be controlled in a socially compliant manner.
2. **Constructing an optimal control problem**: By maximizing the utility functions of the autonomous vehicle and its following vehicles, this problem is formulated as an optimal control problem and solved using Pontryagin's Minimum Principle, ensuring optimality.
3. **Application in eco - driving**: Through a series of numerical simulations, the mechanism and effectiveness of the proposed method in actual traffic data are demonstrated, especially the measured data on Highway 55.

### Formula explanation
- **Utility function**:
\[
J_1=\cos(\phi_{\text{AV}}) U_{\text{AV}}+\sin(\phi_{\text{AV}}) U_{\text{follower}}
\]
where \( U_{\text{AV}} \) and \( U_{\text{follower}} \) represent the "self - benefit" and "following vehicle benefit" of the autonomous vehicle and its following vehicle respectively, and \(\phi_{\text{AV}}\) is the social value orientation angle of the autonomous vehicle.

- **Optimization problem**:
\[
\max J_1 = \max \left( \cos(\phi_{\text{AV}}) U_{\text{AV}}+\sin(\phi_{\text{AV}}) U_{\text{follower}} \right)
\]
under a series of operational constraints, such as the control constraints of the acceleration input variable \( u_i \).

- **Specific form of the utility function**:
- Utility function of the autonomous vehicle \( U_{\text{AV}} \):
\[
U_{\text{AV}}=-\int_I \frac{1}{2} \dot{v}_i(t)^2 \, dt
\]
- Utility function of the following vehicle \( U_{\text{follower}} \):
\[
U_{\text{follower}}=-\int_I \frac{1}{2} [v_{i + 1}(t)-v_0]^2 \, dt
\]

- **Introducing soft state constraints**:
\[
s_i = s_d
\]
where \( s_i \) represents the distance between the autonomous vehicle and the vehicle in front of it, and \( s_d \) is the desired distance.

- **Objective function**:
\[
J_3=\int_I \left( \cos(\phi_{\text{AV}}) \frac{1}{2} \dot{v}_i(t)^2+\sin(\phi_{\text{AV}}) \frac{1}{2} [v_{i + 1}(t)-v_0]^2+\lambda [s_i(t)-s_d]^2 \right) \, dt
\]

### Conclusion
Through the above methods, the paper...