Autonomous Vehicle Lane Change Decision and Control Using Model Predictive Control and Long-Short Term Memory Trajectory Prediction

מושגי ליבה

This paper introduces a decision-making and control method for autonomous vehicle lane change on highways based on Model Predictive Control (MPC). The approach divides the driving control into lane-change decision and lane-change control, both of which are solved using MPC. A Long-Short Term Memory (LSTM) model is used to predict the trajectories of surrounding vehicles to enhance the MPC decision and control modules.

תקציר

The paper presents a two-stage control strategy for autonomous vehicle lane change on highways.
In the lane-change decision module, the minimum driving costs for each lane are computed and compared by solving an MPC problem to determine whether to change lanes or stay in the current lane. The cost function considers factors such as velocity, jerk, distance to leading and following vehicles. A threshold is used to avoid unnecessary frequent lane changes.
The lane-change control module uses a dynamic bicycle model and a multi-objective cost function to obtain the optimal control inputs (acceleration and steering angle) for the lane-change maneuver. Constraints are imposed on velocity, lateral position, steering angle, and acceleration to ensure safe and smooth driving.
A Long-Short Term Memory (LSTM) model with convolutional social pooling is used to predict the trajectories of surrounding vehicles. This predicted information is utilized in both the decision-making and control modules to enhance the performance.
The proposed approach is validated through simulation experiments using the SUMO platform. The results demonstrate that the autonomous vehicle equipped with the proposed control method can effectively identify and navigate suitable paths, with the LSTM-based prediction improving the driving performance compared to a simpler prediction method.

סטטיסטיקה

The ego vehicle maintains a reference speed of 27 m/s.
The time headway during car-following is set to 1.5 s.
The standstill reference following distance is 5 m.
The weight term for jerk is 0.1.
The weight term for leading vehicle distance is 1.
The weight term for following vehicle distance is 0.2.
The lane-change decision threshold is 0.3.
The penalty term on lane change is 0.1.

ציטוטים

"To enable autonomous vehicles to perform discretionary lane change amidst the random traffic flow on highways, this paper introduces a decision-making and control method for vehicle lane change based on Model Predictive Control (MPC)."
"This approach divides the driving control of vehicles on highways into two parts: lane-change decision and lane-change control, both of which are solved using the MPC method."
"A long-short term memory (LSTM) model is used to predict the trajectories of surrounding vehicles for both the MPC decision and control modules."

תובנות מפתח מזוקקות מ:

Highway Discretionary Lane-change Decision and Control Using Model Predictive Control

by Zishun Zheng... ב- arxiv.org 04-04-2024

https://arxiv.org/pdf/2402.17524.pdf

שאלות מעמיקות

How can the proposed approach be extended to handle more complex highway scenarios, such as merging, exiting, or multi-lane changes

To extend the proposed approach to handle more complex highway scenarios like merging, exiting, or multi-lane changes, several enhancements can be implemented. For merging scenarios, the system can incorporate additional sensors to detect merging vehicles and adjust the lane-change decision criteria accordingly. Advanced algorithms can be developed to predict the behavior of merging vehicles and optimize lane changes to ensure smooth merging maneuvers.
For exiting scenarios, the system can utilize GPS data and mapping information to identify exit lanes and plan lane changes in advance. By integrating real-time traffic data and predictive modeling, the system can optimize lane changes to facilitate safe and efficient exits.
In the case of multi-lane changes, the framework can be expanded to consider multiple adjacent lanes simultaneously. By incorporating more complex decision-making algorithms and predictive models, the system can analyze various lane change options and prioritize them based on safety and efficiency criteria. Advanced control strategies can be implemented to coordinate multi-lane changes and ensure smooth transitions between lanes.

What are the potential challenges in implementing this framework on real autonomous vehicles, and how can they be addressed

Implementing this framework on real autonomous vehicles may face several challenges. One major challenge is the integration of the proposed decision-making and control algorithms with the vehicle's existing hardware and software systems. Ensuring seamless communication between sensors, actuators, and the control algorithm is crucial for the system's reliability and performance.
Another challenge is the real-time processing and execution of lane change decisions. The system must be capable of processing large amounts of data, predicting vehicle trajectories accurately, and making split-second decisions to ensure safe and efficient lane changes.
Addressing these challenges requires rigorous testing and validation of the system in various real-world scenarios. Conducting extensive simulations, field tests, and validation studies can help identify potential issues and refine the system's algorithms and parameters. Additionally, continuous monitoring and feedback mechanisms can be implemented to improve the system's performance over time.

How can the lane change decision and control be further optimized to balance safety, efficiency, and passenger comfort

To further optimize lane change decision and control for autonomous vehicles, a holistic approach that balances safety, efficiency, and passenger comfort is essential. One way to achieve this is by integrating advanced AI algorithms, such as reinforcement learning, to continuously learn and adapt to changing traffic conditions. By leveraging real-time data and feedback, the system can dynamically adjust its decision-making process to prioritize safety while optimizing for efficiency and passenger comfort.
Moreover, incorporating human-like driving behavior models into the system can enhance passenger comfort during lane changes. By mimicking human drivers' natural tendencies and preferences, the system can make smoother and more predictable lane change maneuvers, reducing passenger discomfort and enhancing the overall driving experience.
Furthermore, optimizing the control algorithms to consider not only the ego vehicle but also the interactions with surrounding vehicles can improve safety and efficiency. By implementing cooperative behavior prediction and advanced trajectory planning techniques, the system can anticipate and respond to potential conflicts or obstacles during lane changes, ensuring a seamless and safe driving experience for passengers.