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

TL;DR

This work tackles efficient exploration in unknown cluttered indoor environments by leveraging floor-plan prediction. It introduces FPUNet, a UNet++-based predictor with anisotropic convolutions to forecast floor-plan occupancy from local noisy maps, followed by room segmentation and a room-topology graph to guide exploration. A soft, topology-aware exploration strategy combines predicted information with real-time observations by optimizing a visiting order via a linear-programming formulation and a Lin-Kernighan heuristic, expressed as $C(f_i)=\exp(\lambda u_j) C_N(f_i,p(t))$. Across simulations and real-world noisy data, the approach yields substantial path-length reductions (up to around 34% in large scenes) and demonstrates robust generalization, indicating practical potential for scalable autonomous exploration. FPUNet thus enables robust, global-guidance in exploration despite imperfect predictions, advancing efficient map reconstruction in real-world cluttered environments.

Abstract

Robot exploration aims at the reconstruction of unknown environments, and it is important to achieve it with shorter paths. Traditional methods focus on optimizing the visiting order of frontiers 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 to predict is weak. Moreover, this inaccuracy will lead to limited improvement in exploration. Therefore, we propose FPUNet which can be efficient in predicting the layout of noisy indoor environments. Then, we extract the segmentation of rooms and construct their topological connectivity based on the predicted map. The visiting order of these predicted rooms is optimized which can provide high-level guidance for exploration. The FPUNet is compared with other network architectures which demonstrates it is the SOTA method for this task. Extensive experiments in simulations show that our method can shorten the path length by 2.18% to 34.60% compared to the baselines.

Figures (6)

• Figure 1: The bottom left shows the ground truth of the environment, while the right side displays the exploration status at a specific moment. We first obtain the predicted floor plan using FPUNet. Then, the segmentation of rooms (purple area) and their topological connectivity (black line) are extracted to accelerate exploration.
• Figure 2: Illustration of the framework of the proposed method. For real-world scenes, there will be various obstacles in the 2D grid map. To achieve efficient map prediction, we denoise the 2D map first and then perform prediction based on the floor plan. Finally, the predicted map can be used to extract room segmentation and their topology, which provides guidance for the downstream tasks.
• Figure 3: (a) Obtained poses of rooms and doors. The red point represent the poses of rooms. The blue points represent the poses of doors. (b) Created segmentation of the scene. The red and blue areas represent different rooms and doors respectively.
• Figure 4: Illustration for map prediction in different scenes. Grid map: observations; Masks: predicted maps. The first picture in each group is the predicted floor plan. The second picture is the ground truth scene. The scene is getting larger from (a) to (d).
• Figure 5: (a) The influence of obstacle weight for map prediction. (b) The influence $\lambda$ for exploration path length.