Autonomous Object Exploration with Semantics-Aware Routes-Enhanced Next-Best-View Planning

In modern industries, real-time scene understanding has emphasized incorporating semantic information into robotic tasks, particularly in autonomous target searching and transportation. This work introduces a Semantics-Aware Routes-enhanced Next-Best-View Planner i.e., SAR-NBVP designed to explore objects efficiently. The SAR-NBVP consists of a semantic perception strategy and a semantics-aware planning method. In the semantic perception phase, semantic point clouds are generated using RGBD images, which are then mapped into a volumetric map. This process enables the representation of semantic object information within the constructed map. The semantics-aware planning phase considers kinematic, geometric, and semantic costs within the SAR-NBVP. This ensures that the planner effectively searches for semantic objects by considering the robot’s physical constraints as well as the environment’s geometric and semantic characteristics. The proposed perception and planning algorithm’s effectiveness is validated through simulation experiments, demonstrating its ability to incorporate semantic information, efficiently explore the environment, and search objects. The video of the simulation experiment is shown at https://youtu.be/STjwV0xQcvw.