The task of moving a swarm of drones in a limitedspace without preliminary markings can find practical application, both in rescue operations and in the creation of autonomous driving systems for unmanned vehicles. To implement a safe movement pattern for several robots, two main problems must be solved. First, when driving in an unfamiliar environment, the drone must make a decision to change its trajectory in near real time. Secondly, movement is complicated by the presence of moving "brothers" around the robot, which significantly increases the likelihood of collision.
To solve these problems in the GLAS algorithmartificial intelligence is used, which allows the robot to not have a complete and comprehensive picture of the environment. The study of space by robots is carried out in real time, allowing the drone to independently "think" and evaluate the surrounding space. To track the mass of flying robots, a Neural-Swarm swarm tracking controller was also used, which estimates the mutual aerodynamic effect of drones.
When testing a swarm of 16 robots,it was found that the Neural-Swarm and GLAS complex improves flight coordination by 20% compared to traditional methods of planning mass movements of autonomous devices.