Design of an active exploration strategy to support semantic SLAM

The two main elements a mobile robot requires to work autonomously are: The ability to perceive the environment, and the ability to perform an action based on this environment. When a mobile robot acts in an unknown environment, the first element is tackled by an approach called SLAM, or Simultaneous Localization And Mapping, which concerns with the problem of creating a map of an unknown environment by a mobile robot while simultaneously localizing itself in and navigating said map.

More recent algorithms make use of active SLAM, where, as opposed to just passively building a map based on sensor data, the robot also computes appropriate control actions to reduce map and localization uncertainty, and to maximize map coverage. This is also called exploration, which aims to reduce the amount of unseen area on the map.

Because of the recent advancements in low-cost processing power, an emerging SLAM approach is semantic SLAM. Semantic SLAM focuses on getting more information about the contents of the map, using image processing to recognize objects, as opposed to solely creating feature- or filter-based maps like traditional SLAM approaches do. The locations of these objects can then be used to increase the map and localization accuracy. Another advantage is that the mobile robot can use the information about objects for navigation or even to manipulate the environment.

It is very beneficial for the mobile robot to map as may objects as possible to gather the most information into the map so the most informed decisions can be made. In order to do this autonomously, an exploration strategy can be designed to optimize the amount of semantic information gain within a certain timeframe. This can be used to support the semantic slam algorithm to make it faster and more efficient.

This is why the main goal of this research can be defined as follows: Design an active exploration strategy to support the objectives of a semantic SLAM algorithm. This exploration strategy will attempt to optimize the amount of semantic information gain, meaning to map as many objects as fast as possible while still performing the objectives set out by the SLAM algorithm.