Title:P2 Explore: Efficient Exploration in Unknown Clustered Environment with Floor Plan Prediction

Abstract:Robot exploration aims at constructing unknown environments and it is important to achieve it with shorter paths. Traditional methods focus on optimizing the visiting order based on current observations, which may lead to local-minimal results. Recently, by predicting the structure of the unseen environment, the exploration efficiency can be further improved. However, in a cluttered environment, due to the randomness of obstacles, the ability for prediction is limited. Therefore, to solve this problem, we propose a map prediction algorithm that can be efficient in predicting the layout of noisy indoor environments. We focus on the scenario of 2D exploration. First, we perform floor plan extraction by denoising the cluttered map using deep learning. Then, we use a floor plan-based algorithm to improve the prediction accuracy. Additionally, we extract the segmentation of rooms and construct their connectivity based on the predicted map, which can be used for downstream tasks. To validate the effectiveness of the proposed method, it is applied to exploration tasks. Extensive experiments show that even in cluttered scenes, our proposed method can benefit efficiency.