insight - Distributed Systems - # Distributed Model Predictive Control for Autonomous Vehicle Platoons

Core Concepts

This paper presents a distributed model predictive control (DMPC) algorithm for heterogeneous vehicle platoons that can accommodate affine spacing policies, such as constant distance or constant time headway, and arbitrary communication topologies where each vehicle can communicate with at least one preceding vehicle.

Abstract

The paper proposes a DMPC algorithm for controlling heterogeneous vehicle platoons. The key highlights are:
The algorithm can handle any affine spacing policy, including constant distance headway (CDH) and constant time headway (CTH), between vehicles.
The algorithm can work with arbitrary communication topologies, as long as each vehicle can communicate with at least one preceding vehicle in the platoon.
A stability analysis is provided, deriving a sufficient condition on the weights in the DMPC cost function to guarantee asymptotic stability of the platoon.
Simulation experiments with a platoon of 50 vehicles and hardware experiments with a platoon of four 1/10th scale vehicles are conducted to validate the algorithm and compare the performance under different spacing policies and communication topologies.
The results show that the CTH policy generally outperforms the CDH policy in terms of spacing error, especially for vehicles further down the platoon, due to the ability to better handle sudden changes in the leader's velocity.

Stats

The paper presents the following key figures and metrics:
The platoon consists of N+1 vehicles, where vehicle 0 is the virtual leader and vehicles 1 to N are the followers.
The simulation experiments use a platoon of 50 vehicles (N=50).
The hardware experiments use a platoon of 4 vehicles.
The discrete timestep used in the experiments is 0.1 seconds.
The prediction horizon used in the DMPC algorithm is 6 seconds (H=60).
The desired distance for the CDH policy is 5 meters for vehicles 2 to N, and 0 meters for vehicle 1.
The time headway for the CTH policy is 0.2 seconds, with a safety distance of 1 meter for vehicles 2 to N, and 0 seconds with 0 meters for vehicle 1.

Quotes

"The platooning problem can be decomposed into four components [9]: vehicle dynamics, communication topology, control algorithms, and spacing policies."
"It is widely known that when using linear feedback controllers with a PF topology, a constant distance headway (CDH) spacing policy (each vehicle maintains some predefined distance to its predecessor) leads to string instability [10], i.e., errors seen by the leader propagate and worsen down the platoon. A constant time headway (CTH) policy (vehicles remain some set time behind the predecessor) alleviates this issue [11], [12]."

Key Insights Distilled From

by Michael H. S... at **arxiv.org** 04-22-2024

Deeper Inquiries

In a practical deployment scenario where communication delays or packet losses are common, the DMPC algorithm can be extended by incorporating predictive models and robust control techniques. One approach is to include prediction models that account for communication delays, allowing the algorithm to anticipate future states and adjust control inputs accordingly. This predictive capability can help mitigate the impact of delays on the platoon's performance.
Additionally, the algorithm can be enhanced with robust control strategies that are resilient to packet losses. By incorporating redundancy in communication channels or implementing error correction mechanisms, the algorithm can ensure reliable data transmission even in the presence of packet losses. Robust control techniques can also help the platoon adapt to uncertainties in communication, maintaining stability and performance under varying conditions.

As the platoon size increases, the performance of the DMPC algorithm may improve in terms of stability, accuracy, and efficiency in maintaining desired platoon formations. However, this improvement in performance may come at the cost of increased computational complexity.
The main trade-offs between performance and computational complexity as the platoon size grows include:
Increased Communication Overhead: With a larger platoon, the amount of communication between vehicles also increases, leading to higher communication overhead. This can result in more data to process and transmit, impacting computational complexity.
Higher Control Computation: Managing a larger number of vehicles in the platoon requires more complex control computations to optimize trajectories, spacing, and coordination. This can lead to increased computational load and processing time.
Resource Allocation: As the platoon size grows, resource allocation becomes more challenging. Balancing the computational resources needed for real-time control decisions with the increasing number of vehicles in the platoon can be a trade-off between performance and complexity.
Scalability: The algorithm's scalability in handling a larger platoon size while maintaining real-time responsiveness and stability is a critical trade-off. Ensuring that the algorithm can scale effectively without sacrificing performance requires careful consideration of computational demands.

To handle heterogeneous vehicle dynamics and spacing policies within the same platoon, the DMPC algorithm can be adapted in the following ways:
Dynamic Model Formulation: Develop a dynamic model that can accommodate different vehicle dynamics within the platoon. This model should be flexible enough to capture variations in vehicle parameters such as mass, inertia, and control constraints.
Parameterized Cost Function: Modify the cost function in the DMPC algorithm to account for heterogeneous dynamics. Introduce parameters that capture the differences in vehicle characteristics and spacing policies, allowing the algorithm to optimize control inputs based on individual vehicle requirements.
Adaptive Control Strategies: Implement adaptive control strategies that can adjust control actions based on the specific dynamics and spacing policies of each vehicle. This adaptive approach enables the algorithm to tailor control decisions to the unique characteristics of each vehicle in the platoon.
Communication Protocols: Enhance the communication protocols to exchange information about individual vehicle dynamics and spacing policies. By sharing this data within the platoon, vehicles can coordinate effectively and maintain desired formations while accommodating heterogeneity.
Multi-Objective Optimization: Consider multi-objective optimization techniques to optimize control inputs while balancing conflicting objectives arising from heterogeneous dynamics and spacing policies. This approach can help achieve a trade-off between different performance metrics for diverse vehicles in the platoon.

0