ASEP: An Autonomous Semantic Exploration Planner With Object Labeling

Abstract

In this paper, we present a novel 3D autonomous exploration planner called the Autonomous Semantic Exploration Planner (ASEP), designed for GPS-denied indoor environments. ASEP combines real-time mapping, exploration, navigation, object detection, and object labeling onboard an Unmanned Aerial Vehicle (UAV) with limited resources. The planner is based on a frontier exploration strategy that utilizes semantic information about the environment in the exploration policy. The policy is extended to incorporate both geometric and semantic information provided by a deep convolutional neural network (DCNN) for semantic segmentation. This semantically-enhanced exploration algorithm directs the exploration toward the quick labeling of all objects of interest in the environment. An extended path planning algorithm continuously checks for path validity, enabling safe navigation in challenging environments. The overall system is designed to be modular and easily extended or replaced with custom modules. The proposed planner is evaluated and analyzed in both simulation and real-world environments using a UAV. Experimental studies demonstrate the effectiveness of the ASEP strategy compared to state-of-the-art methods. Results show that the objects in the environment are explored faster and total exploration time is reduced while the computational time remains consistent regardless of the semantic segmentation processing involved.