# Fully Distributed Model Predictive Control of Connected Automated Vehicles in Intersections: Theory and Vehicle Experiments

**Authors:** Alexander Katriniok, Benedikt Rosarius, Petri Mähönen  
**arXiv:** [2104.07383](https://arxiv.org/abs/2104.07383)  
**Date:** 2021-04-15  
**Summary page:** https://www.summarizepaper.com/en/paper/2104.07383v2/

## Key points

- Paper proposes a fully distributed control system architecture for coordinating connected and automated vehicles (CAVs) at road intersections
- Aim is to ensure safe and efficient coordination among CAVs
- Control system architecture designed for in-vehicle implementation, allowing each vehicle to make local decisions based on its own perception of the environment
- Approach builds upon a fully distributed model predictive control (MPC) framework
- Each vehicle solves a nonconvex optimal control problem (OCP) locally and synchronously
- Optimized trajectories exchanged among vehicles using vehicle-to-vehicle (V2V) communication
- Penalty convex-concave procedure applied to address computational challenges associated with solving nonconvex OCPs in real-time
- Proposed architecture includes a localization layer responsible for self-localization and an estimator that determines joint collision points with other agents
- Experimental tests demonstrate efficacy of proposed architecture in coordinating CAVs at intersections, ensuring safe and efficient traffic flow

## Summary

The paper titled "Fully Distributed Model Predictive Control of Connected Automated Vehicles in Intersections: Theory and Vehicle Experiments" proposes a fully distributed control system architecture for coordinating connected and automated vehicles (CAVs) at road intersections. The aim is to ensure safe and efficient coordination among CAVs. The control system architecture is designed for in-vehicle implementation, allowing each vehicle to make local decisions based on its own perception of the environment. The approach builds upon a fully distributed model predictive control (MPC) framework, where each vehicle solves a nonconvex optimal control problem (OCP) locally and synchronously. The optimized trajectories are then exchanged among vehicles using vehicle-to-vehicle (V2V) communication. To address the computational challenges associated with solving nonconvex OCPs in real-time, the authors apply the penalty convex-concave procedure. This procedure involves solving a convexified version of the original OCP, enabling faster computation while still maintaining satisfactory performance. In addition to the predictive controller, the proposed architecture includes a localization layer responsible for self-localization and an estimator that determines joint collision points with other agents. These components enhance the overall effectiveness of the control system. The efficacy of the proposed architecture is evaluated through experimental tests which demonstrate that it effectively coordinates CAVs at intersections, ensuring safe and efficient traffic flow. Overall, this paper presents a comprehensive approach to addressing coordination challenges in CAVs at intersections by leveraging fully distributed MPC and V2V communication along with localization and collision estimation techniques; thus offering a promising solution for enhancing safety and efficiency in future transportation systems.