Sürü robotların müşterek hareketinde beklenti

Recent technological advancements made the implementation of swarms of UAVs possible. As they face more complex situations each day, collective motion models rise in importance. In this thesis, a collective motion model for self-propelled agents with anticipative action is given. It is shown that using anticipated positions in the attraction-repulsion mechanism brings a new interaction term that depends on velocities. The noise-induced order-disorder phase transition of this model is compared with two other well-known collective motion models: Active-Elastic model and Vicsek model. The comparison shows that anticipation aligns the headings of the agents. Therefore, the anticipation horizon can be adjusted to help robot swarms accomplish certain tasks. For example; better braking performance and preventing pileup in case of a sudden stop, or squeezing through a narrow passage without losing order. Sensors used in the robot swarms are mostly position-based. Since velocity information is required for the anticipative model, a Kalman Filter that takes equations of motion into consideration is designed to improve the position measurements and supply velocity and heading information. Particle-based simulations are done for the proposed model with the designed filter. Then, the model and the filter are implemented in the Crazyswarm platform, which is used for flying a swarm of Bitcraze Crazyflie 2.x quadcopter UAVs. In the experiments, control commands of the agents are created by their own Kalman Filter, which uses noisy measurements and the motion model information. Experiments with the real UAVs show that the swarm is able to pass through a narrow passage with the proposed anticipative motion model.