Intelligent robot systems are gradually moving from a single-machine-centered task execution mode to a multi-robot system with more collaborative capabilities. Especially in the cluster formation scenario of unmanned vehicles and UAVs, how to achieve efficient and stable collaborative operations has become a hot topic in robotic system research and engineering deployment. This article focuses on the vehicle-UAV coordination cluster formation kit launched by AMOVLAB, introduces its implementation ideas in communication architecture and control methods, and discusses the implementation route of vehicle-UAV coordination formation.

01 cluster communication

In multi-robot systems, communication is the core to achieve efficient collaboration. According to different communication equipment and software architecture, it can be divided into the following categories:

Communication equipment network architecture

Centralized: There is a master node, and other slave nodes only communicate with the master node, and data between nodes needs to be transferred through the master node. Common ones include routers, master-slave wireless stations, etc. The advantage is that the architecture is simple, but the disadvantage is that the master node can easily become a performance bottleneck.

distributed: All nodes have equal status and can communicate directly in pairs. They have strong scalability but more complex design.

Hybrid: Some nodes in the cluster are interconnected with each other to form a local distributed network, and other nodes are connected to the master node in a slave manner, taking into account both flexibility and control centralization.

Communication software architecture

Centralized communication system (such as ROS1 multi-unit network):
A host runs ROS Master, and other nodes establish communication with it by configuring IP to realize topic and service interaction. The configuration is simple and easy to deploy, but there is a risk of global naming conflict, and the program needs to run around the master node, resource utilization is limited, and performance depends on the load of the master node.

Distributed communication system:

Distribution requires self-programming. Take AMOVLAB’s cluster communication system as an example.

1. Centerless communication network
The system is based on socketTCP/IP communication mechanism and consists of Prometheus ground control station Together with each unmanned device, it forms a centerless communication network.
2. Bidirectional reliable link (TCP)
Both the ground control station and the unmanned equipment node have created TCP servers and clients. Control instructions are sent to the device from the ground control station through TCP; the device returns heartbeat packets through TCP to implement status monitoring and connection maintenance.
3. Real-time status channel (UDP)
The ground control station receives data as a UDP client. The device not only acts as a UDP server, but also actively broadcasts: - unicast method to send task status information to the ground control station; - multicast method to broadcast some topic messages to the LAN, and other nodes only need to subscribe to the multicast address to share information.
4. Custom protocol format
Encode, parse and route all messages to ensure transmission flexibility and system scalability.

02 Cluster control

The control system is the core of the operation of multi-robot clusters. Appropriate collaborative control strategies need to be flexibly selected in different scenarios to ensure the overall collaborative effect.

lead-follow method

Designate one UAV as the leader, and the other UAVs (Followers) adjust their status according to the leader's position and speed and maintain the preset relative positions.Applicable scenarios: Collaborative operations with a relatively fixed structure and known task paths, such as cruising in formation, passing in narrow areas, etc. The algorithm has simple calculation, low communication overhead and high stability.

virtual structure method

The entire UAVformation is regarded as a rigid virtual structure. Each UAV corresponds to a fixed point on the structure, and the formation is adjusted by controlling the movement of the virtual structure.Applicable scenarios: Suitable for tasks that require stable maintenance of complex formation structures, such as synchronized flight, formation display, multi-point perception collaboration, etc. It has strong scalability and is conducive to form switching.

behavioral law

Define a variety of behavior rules (such as obstacle avoidance, following, cruising), and the UAV dynamically adjusts its behavior priority according to the environment.Applicable scenarios: Scenarios with frequent task changes and high environmental uncertainty, such as inspections, search and rescue, exploration tasks, etc. System responses are more adaptive, but behavioral conflicts require additional handling.

artificial potential field method

Adjust UAV motion by designing attractive potential fields (target points) and repulsive potential fields (obstacles, other UAVs).Applicable scenarios: It is suitable for complex navigation tasks in space that require simultaneous obstacle avoidance and trend targets, such as multi-machine collision avoidance, autonomous localization, and multi-point coverage tasks. The method is intuitive, but there may be a trapping problem at local minima.

03 AMOVLABvehicle-UAV coordinationformation kit

With the support of a centerless communication network, formation control is responsible for converting "multi-node perception" into "coordinated motion". AMOVLAB’s vehicle-UAV coordinationformation suite supports two mainstream collaborative control methods:Lead-Follow Method and virtual structure method, Can be dynamically switched according to task complexity and environment.

How to implement the leader-follower method

In this method, it is assumed that the position of each follower in the leader's own coordinate system remains unchanged. As the leader moves along the preset trajectory, the follower obtains its status information in real time, converts its coordinates into the ENU system (East-North-Up), and then superimposes its relative offset to generate the target position, thereby achieving dynamic formation control. This method has simple calculation and low communication overhead, and is suitable for scenarios where the formation structure is relatively fixed.

// Rotation matrix: body frame to inertial frame  
    R_Body_to_ENU = get_rotation_matrix(this->uav_states_[0].attitude[0],this->uav_states_[0].attitude[1],
                                                                            this->uav_states_[0].attitude[2]);
    this->pos_enu_frame =  R_Body_to_ENU * this->pos_body_frame;
    if(this->follow_start_flag_)
    {
        if(this->agent_id_ == 1)
        {
            this->follow_pos[0] = this->all_uav_pos_[0][0];
            this->follow_pos[1] = this->all_uav_pos_[0][1];
            this->follow_pos[2] = this->all_uav_pos_[0][2];
        }else
        {      
            this->follow_pos[0] = this->all_uav_pos_[0][0] + this->pos_enu_frame[0];           
            this->follow_pos[1] = this->all_uav_pos_[0][1] + this->pos_enu_frame[1];                              
            this->follow_pos[2] = this->all_uav_pos_[0][2] + this->pos_enu_frame[2];
        }
    }

Implementation method of virtual structure method

The virtual structure method abstracts the entire formation into a rigid structure, in which each device corresponds to a fixed point. The system supports common formations such as straight lines, triangles, squares, circles, etc., and automatically generates the coordinates of each point based on the input device number and spacing parameters. Through the allocation algorithm, the system matches the nearest target point for each device and sends corresponding control instructions to guide it into position and complete the formation reconstruction. This method is suitable for complex scenarios of dynamic transformation of formation shapes or multi-device switching tasks.