Multi Agent Setup

In order to coordinate the exploration or inspection by multiple agents, collisions have to be avoided and the agents have to be aware, when their field of view is blocked by their peers. To this end this repository contains the two packages multiagent_collision_checker and pointcloud_pruner. The first is used to make sure that the agents do not collide, while the latter prunes the incoming pointcloud, such that no points representing peer agents are inserted in the occupancy map. More functionality is added to the nbvplanner package to allow the detection of other agents in the field of view when the inspected surface area is updated.

In the following, an overview is provided on how the different packages become aware of the relative locations of the peer agents. This is especially critical with real robots, when no common ground truth is available.

Rotors Simulation

Component	Channel
Collision check in tree construction	Publish next segments to topic /evadeSegment for all peers (using multiagent_collision_checker
Occlusion check of the pointcloud to be integrated in the octomap	pointcloud_pruner listens to the TFs of peers and prunes the incoming pointclouds accoridingly
Occlusion check of the surface gain computation	Not checked, as all agents move and prediction is difficult and unnecessary
Occlusion check of the update of inspected surface	Planner listens to pose topic of all agents and discards update if field of view is not free

ASL Firefly/Neo Setup

TBD