Multiagent System for smart meta-sensor framework

Multiagent system for smart meta-sensor framework In Industry 4.0 the AMRs (Autonomous Mobile Robots) have a key role. The AMRs need to share the same spaces with humans but only some of the AMRs can detect and recognize them through specific sensor data. Some types of sensors perform better than others in this context, but they might increase the cost of those robots. Robots equipped with only LIDAR (Light Detection And Ranging) perform well for indoor SLAM but still have limitations regarding obstacles that are below the height of the LIDAR. However, if they are also equipped with an RGB-D sensor, a more detailed view enriches the robot's perception. ROS (Robot Operating System) represents the state-of-the-art framework for the coordination and management of robots. ROS offers a Navigation Stack commonly used by the majority of AMRs. The ROS Navigation Stack is fully modular, and every manufacturer tunes this package depending on the characteristics of its own robot. The key idea for this thesis is to create a ROS framework to share information between different types of robots, allowing them to assist each other in the early identification of obstacles. This will enable robots with limited vision capabilities to perceive obstacles that they would not be able to detect on their own. The main information that needs to be shared is the robot's position and the dynamic obstacles independently of the implementation and algorithms used by the various robots. In this framework, each robot shares its real-time position with a node called robot_info_manager_node, which is responsible for forwarding this information to all other robots. This ensures that each robot can consider the positions of others during the path-planning phases. This node not only receives and shares the positions of all the robots but also manages their local_costmap. These costmaps represent the real-time dynamic view of each robot and what its sensors perceive. To do this, each robot is registered with this node through a configuration file, and each robot is assigned a name. Using a service, each robot can send an information request to the node, specifying who is making the request. The node will respond by sending all the local\_costmap and the positions of the other robots, excluding the requesting robot itself. The service is called through a new layer global\_costmap called positioning\_layer, which all robots that want to receive information from other robots must add to the list of layers in the global\_costmap, using the configuration file of the move\_base node. This layer will be responsible for marking cells as occupied where various robots are positioned to avoid collisions between robots, and for copying their local_costmap to the correct position, including obstacles that, for structural reasons, cannot be observed. The only constraints are that the robots share the same map and the resolution of local_costmap and global_costmap are the same. This framework has been tested using a LoCoBot WX250S from Trossen Robotics and a Turtlebot3 Burger from Robotis, both in simulation using Gazebo and in a real-world environment in a laboratory.