Publicação
Robot Navigation in Highly-Dynamic Environments
| Resumo: | Autonomous mobile robots have evolved fromsingle purpose machines used in isolated, custom built robot-specificenvironments, to complex multi-purpose machines interacting withhumans in their natural environments. The presence of such robots has also increasedin many application areas, with industrial robots in factories and service robotsin public areas becoming more and more important and accepted. These environments, where the robot may find obstacles in unexpected locations and need to replan its path in accordance with theirmotion, are considered dynamic environments. Is is, therefore, important to assure that robots are capable of navigating in a safe and collision-free manner in dynamic environments through adequate path planning that takesinto consideration the motion of the obstacles.This dissertation presents a novel dynamic obstacle avoidance approach, aiming to avoid particularly humans, taking into account their future poses as predicted by their motion. The approach generates an occupancy map, represented as a costmap, in accordance with the future poses, improves it with social constraints, and uses A* to find a collision free and socially acceptable path. However, the approach is not limited to A*, it can use other path planning algorithms when required.We take into account the motion ofthe obstacles to predict the area that the robot should avoid goingthrough in the future in order to prevent a collision, instead of assuming a staticenvironment. Our approach also incorporates the uncertainty in theestimate of the obstacles’ motion into the costs assigned to thecost-map. We ensure that the generated paths do not hinder theobstacle’s motion and lead the robot through a socially more acceptable trajectory by inflating the costs of the costmap cells located along theobstacle’s moving direction.The proposed solution was validated through both simulation and real-world experiments wherecollisions would be bound to happen unless the robot had the ability to avoid them. Results show that cost assignmenttechniques that do not account for the motion of the obstacles are not reliable for navigation in dynamic environments.In contrast, our experiments revealed our approach to be able to, on most cases, avoid collisions and lead the robot to perform more socially acceptable trajectories by passing the obstacles through their backs when appropriate. |
|---|---|
| Autores principais: | Silva, Carlos André Seara da |
| Assunto: | Evitação de obstáculos dinâmicos Navegação social Predição de colisões Robôs móveis Planeamento de trajetória Dynamic obstacle avoidance Social navigation Collision prediction Mobile Robot Path Planning |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Coimbra |
| Idioma: | inglês |
| Origem: | Estudo Geral - Universidade de Coimbra |
| Resumo: | Autonomous mobile robots have evolved fromsingle purpose machines used in isolated, custom built robot-specificenvironments, to complex multi-purpose machines interacting withhumans in their natural environments. The presence of such robots has also increasedin many application areas, with industrial robots in factories and service robotsin public areas becoming more and more important and accepted. These environments, where the robot may find obstacles in unexpected locations and need to replan its path in accordance with theirmotion, are considered dynamic environments. Is is, therefore, important to assure that robots are capable of navigating in a safe and collision-free manner in dynamic environments through adequate path planning that takesinto consideration the motion of the obstacles.This dissertation presents a novel dynamic obstacle avoidance approach, aiming to avoid particularly humans, taking into account their future poses as predicted by their motion. The approach generates an occupancy map, represented as a costmap, in accordance with the future poses, improves it with social constraints, and uses A* to find a collision free and socially acceptable path. However, the approach is not limited to A*, it can use other path planning algorithms when required.We take into account the motion ofthe obstacles to predict the area that the robot should avoid goingthrough in the future in order to prevent a collision, instead of assuming a staticenvironment. Our approach also incorporates the uncertainty in theestimate of the obstacles’ motion into the costs assigned to thecost-map. We ensure that the generated paths do not hinder theobstacle’s motion and lead the robot through a socially more acceptable trajectory by inflating the costs of the costmap cells located along theobstacle’s moving direction.The proposed solution was validated through both simulation and real-world experiments wherecollisions would be bound to happen unless the robot had the ability to avoid them. Results show that cost assignmenttechniques that do not account for the motion of the obstacles are not reliable for navigation in dynamic environments.In contrast, our experiments revealed our approach to be able to, on most cases, avoid collisions and lead the robot to perform more socially acceptable trajectories by passing the obstacles through their backs when appropriate. |
|---|