Path Planning for Active V-Slam Based on Reinforcement Learning

Slam (Simultaneous Localization and Mapping) is a passive system and in traditional slam algorithm robot’s path is not considered when improving localization uncertainty. However, improving localization accuracy while autonomously exploring unknown environments needs to get abundant feature points and make enough loop closures. To that end we propose a reinforcement learning based active slam framework that can add path planning to existing slam algorithms. In this framework a reinforcement learning agent plans the path while slam is processing. We have tested our framework in simulation environments built in Unreal engine with unrealcv plugin and we have got excellent results.