"Chain Flocking: Multi-Agent Navigation in Unknown Environments with Local Measurements"

Recently, Multi-Agent Systems (MAS) have gained significant importance in various robotics applications due to their ability to cooperate and distribute tasks efficiently. This research presents a novel approach to navigating a MAS in unknown environments without the use of a global positioning system, a common scenario in many missions. Effective navigation is essential as it forms the foundation for executing any subsequent tasks.
The proposed solution is based on a flocking scheme, eliminating the need for predefined formation geometry and enabling the formation scalability and adaptability. In this approach, agents perceive relative information about other agents and obstacles to maintain the formation and avoid collisions. Additionally, agents can perceive relative information about the goal location and navigate the environment, either by using local motion planning techniques or by following each other to maintain the formation, resulting in the creation of a chain structure within the overall flocking formation.
This behaviour, referred to as Chain Flocking, is a multi-leader-follower formation scheme where each agent can lead another agent while following the primary formation leader to navigate the environment. To achieve a fully distributed formation without predefined leader, a distributed method is proposed where a cost function is used to select the overall formation leader, who utilizes local motion planning for navigation. This leader can change based on the formation’s needs, depending on the agent’s location, goal information and environment characteristics. By having agents follow one another, the formation topology is modified to increase the agent’s perception range without compromising the integrity of the flocking formation geometry.
The effectiveness of the method is demonstrated through multiple simulations using a dynamic simulator and differential-drive robots as agents. In the experiments, multiple robots with the Chain Flocking algorithm implemented are placed in different environments and successfully navigate from one origin location to another using relative sensing information about other agents, obstacles and the goal.

"Planificación de Movimiento con Retroalimentación Sensorial"

En esta tesis se propone un método que integra en el planificador local retroalimentación sensorial, más en especifico se propone extender los mapas de caminos probabilísticos (PRMs por sus siglas en Inglés) incluyendo control visual basado en imagen (IBVS por sus siglas en Inglés) en el planificador local. El objetivo es complementar las fortalezas de ambas técnicas, obteniendo un método que por construcción es capaz de considerar obstáculos y ser ejecutable a lazo cerrado con control visual. A diferencia de trabajos previos, el planificador propuesto en esta tesis es multiconsulta y se puede usar para moverse de diversos estados iniciales a diversas estados finales. Se proponen diversas variantes del algoritmo, algunas son capaces de permutar entre varios blancos visuales, y así generar una trayectoria más larga y compleja. Los diferentes algoritmos son validados intensivamente con diversas simulaciones computacionales.