Em busca de navegação socialmente aceita e confortável com foco em cadeiras de rodas autônomas

Abstract

Service robots in public environments are becoming increasingly common. One of the next steps for robotics is the development of autonomous personal navigation robots, such as Autonomous Motorized Wheelchairs (AMW). An AMW differs drasti cally from other robots in terms of navigation, as it must take into account the fact that the robot is actively carrying a person who is exposed to various social interactions, turning the robot into an agent that needs to adhere to social rules during navigation. For example, a traditional autonomous robot is not subject to discomfort during navigation, which can significantly impact the user’s experience in an AMW. This study primarily addresses the dual need for AMWs to navigate efficiently while adhering to social norms, both in interactions with pedestrians and with the wheelchair’s user. The semantics embedded in these social interactions are encoded in layers of costmaps that represent the environment on a two-dimensional grid map, with each cell in this grid holding a value associated with the described semantics. In addition to the traditional cost maps for autonomous robot navigation, we use a layer for third-party personal space, a layer that calculates comfort when traveling in an environment concerning the distance from walls, and a layer that calculates visibility of the location. A recurring problem is the isolated evaluation of these cost layers, without considering the interactions resulting from their combination. The main objective of this work is to analyze how these combined cost maps with different semantics behave concerning both the AMW user and other people. Consequently, two different simulated scenarios were developed, where an AMW must navigate between two points. One scenario reflects a test environment used by a well-established study in the field, while the other approximates a real-world scenario. Thus, the factors evaluated from the simulations were the navigation completion rate, the total distance traveled by the robot, the rate of invasion of humans’ personal space, the comfort of AMW navigation, and the visibility of the user. Regarding the metrics, the combination of all the layers used showed better results in all metrics, except for the total distance traveled. Among the resulting behaviors, for example, it was observed that the inclusion of the comfort and visibility layers had a positive impact on the smoothness coefficient of navigation, thus bringing even more comfort to the user’s navigation. However, in the visibility coefficient results, there was a reduction after the inclusion of the comfort and visibility layers without the personal space layer, highlighting the need to analyze how changes in the gain of each cost function of the layers contribute to the final result of the costmaps.