Path planning is a fundamental issue in robotic systems because it requires coordination between the environment and an agent. The path-planning generator is composed of two modules: perception and planning. The first module scans the environment to determine the location, detect obstacles, estimate objects in motion, and build the planner module’s restrictions. On the other hand, the second module controls the flight of the system. This process is computationally expensive and requires adequate performance to avoid accidents. For this reason, we propose a novel solution to improve conventional robotic systems’ functions, such as systems having a small-capacity battery, a restricted size, and a limited number of sensors, using fewer elements. A navigation dataset was generated through a virtual simulator and a generative adversarial network to connect the virtual and real environments under an end-to-end approach. Furthermore, three path generators were analyzed using deep-learning solutions: a deep convolutional neural network, hierarchical clustering, and an auto-encoder. Since the path generators share a characteristic vector, transfer learning approaches complex problems by using solutions with fewer features, minimizing the costs and optimizing the resources of conventional system architectures, thus improving the limitations with respect to the implementation in embedded devices. Finally, a visualizer applying augmented reality was used to display the path generated by the proposed system.
This is an open access article distributed under the Creative Commons Attribution License
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited