Object-sensitive potential fields for mobile robot navigation and mapping in indoor environments

Mobile robots may be deployed in indoor environments for numerous tasks such as object manipulation, search and rescue, and infrastructure mapping. To be safely deployed in unknown and cluttered environments, mobile robots should be able to recognize both static and dynamic obstacles in its surroundings to determine its navigable paths. Current methods for navigating a mobile robot while avoiding obstacles do not consider the semantic categorization of objects when deciding how to move around them. This study proposes an object-sensitive potential field method for mobile robot navigation and mapping in indoor environments. The mobile robot uses laser scanners to sense the 3D geometry of its surroundings. Next, a point-cloud based object detector is trained using deep learning techniques to semantically parse the laser scan data. The object detection output is used to create potential fields with varying strengths corresponding to the types of objects. This allows the path planning routine to strongly weigh immobile obstacles such as walls while weakly weighing potentially mobile obstacles such as chairs. The proposed algorithm was implemented and tested with a custom-designed mobile robot platform, Ground Robot for Mapping Infrastructure (GRoMI), in an indoor corridor environment. Overall, the proposed algorithm allowed for more intelligent navigation of mobile robots in cluttered environments by taking into account the categorization of obstacles.