A stochastic approach to generate emergent behaviors in robotic swarms

• Autor(a) principal: Paulo Alfredo Frota Rezeck
• Data de Publicação: 2023
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/65401
• topic: Robotics; Multi-agent system; Swarm robotics; Probabilistic robotics; Computação – Teses; Robótica – Teses; Robôs – Sistema de controles – Teses

In this dissertation, we present a novel methodology that extends the concepts of Gibbs Random Fields (GRFs) to the context of swarm robotics, allowing us to design control mechanisms that produce different swarm behaviors using only local information. In this context, a GRF is a probabilistic graph model that describes the interactions and behaviors of a group of robots. The robots are assumed to interact with each other in a way that can be described by a set of random variables. These variables define a random field, and the joint probability function is a Gibbs distribution that describes the probability of the swarm being in a given configuration. By using a Markov Chain Monte Carlo (MCMC) method, each robot sample velocity commands in a decentralized way, forcing them to move toward the global minimum of the potential, leading the entire swarm to converge to the desired behavior. This approach has several advantages over more traditional methods for controlling the behavior of a swarm. For example, it allows for decentralized control, where each robot makes decisions based on local information rather than relying on a central controller. This makes the system more robust and scalable, as there is no single point of failure, and the swarm can continue to operate even if individual robots fail. Additionally, it allows for a more flexible and adaptable approach to swarm behavior, as the potential function can be modified to account for changing environmental conditions or new goals for the swarm. To demonstrate the application of our methodology, we investigate the design of methods that tackle three significant challenges in swarm robotics: flocking and segregation, cooperative object transportation, and pattern formation. Numerical simulations and real-world experiments show that these methods are scalable, adaptable, and robust, even in the presence of noise, failures, and changes in the environment. More specifically, the first method shows to be able to adequately segregate a group of heterogeneous robots while keeping cohesive navigation and avoiding obstacles in the environment. The second method supports the transportation of objects of different shapes, sizes, and masses. It is also scalable and resilient to changes in goal location and robot failures. The third method experiments show the ability to create diverse patterns using different neighborhood constraints and that it may serve as a basis for more tangible applications such as the construction of chain or bridge-like structures using a swarm of heterogeneous robots. Overall, the proposed methodology shows promise for contributing to the field of swarm robotics, enabling the designing of mechanisms that adequately produce different behaviors of a swarm of robots.