Robotic Exploration of Unknown 2D Environment Using a Frontier-based Automatic-Differentiable Information Gain Measure

Abstract

At the heart of path-planning methods for autonomous robotic exploration is a heuristic which encourages exploring unknown regions of the environment. Such heuristics are typically computed using frontier-based or informationtheoretic methods. Frontier-based methods define the information gain of an exploration path as the number of boundary cells, or frontiers, which are visible from the path. However, the discrete and non-differentiable nature of this measure of information gain makes it difficult to optimize using gradientbased methods. In contrast, information-theoretic methods define information gain as the mutual information between the sensor's measurements and the explored map. However, computation of the gradient of mutual information involves finite differencing and is thus computationally expensive. In this work, we propose an exploration planning framework which combines sampling-based path planning and gradient-based path optimization. The main contribution of this framework is a novel reformulation of information gain as a differentiable function. This allows us to simultaneously optimize information gain with other differentiable quality measures, such as smoothness. The effectiveness of the proposed planning framework is verified both in simulation and in hardware experiments using a Turtlebot3 Burger robot.

FLYBO: A Unified Benchmark Environment for Autonomous Flying Robots

“…Despite its appearant simplicity, the method also performs very well in 3D and is still considered revelant as a baseline system in terms of 3D exploration efficiency [8,82]. The seminal work of Yamauchi et al [78] has led to numerous extensions including multi-agent exploration [79], and more sophisticated strategies to extract frontiers using stochastic differential equations [67], information theory and a multi-objective utility function [21], or more generally, in a Next-Best View (NBV) fashion, i.e., by maximizing a given utility function greedilly [24]. Another trend in frontier-based planning is to enhance its main original drawbacks of speed and efficiency.…”

Coordinated Aerial-Ground Robot Exploration via Monte-Carlo View Quality Rendering

“…Next, both agents' goals, q goal = [q uav(goal) , q ugv(goal) ], are selected using the proposed Monte-Carlo view quality renderer (Line 5), which is presented in Sec. V. Then, a sampling-based algorithm like RRT plans collision-free paths, Q, from q 0 to q goal (Line 6), which is further optimized with respect to a frontier-based differentiable information gain measure (Line 7) [13]. Finally, the optimized path, Q * , is executed by the agents to update the occupancy map M (Lines 8-9).…”

“…Our gradient path optimization algorithm adopts the frontierbased automatic-differentiable information gain measure [13] for exploring an unknown 3D environment and increasing visible frontiers along the path. The automatic-differentiable information gain of a viewpoint is achieved by adding a fuzzy logic filter to count visible frontier voxels surrounding a viewpoint so that its gradient is efficiently computed.…”

“…Besides, it avoids inefficient sampling and repetitive computation of sampling-based methods. The paths from the robots' current configurations to the selected goals are further optimized with a frontier-based automaticdifferentiable information gain measure [13]. Compared with paths generated using a single agent, paths generated and optimized with the proposed method are consistently superior in planning time and coverage efficiency.…”

FUEL: Fast UAV Exploration using Incremental Frontier Structure and Hierarchical Planning

“…[4] presented a different scheme of frontier selection to enable higher flight speed, which improves the exploration rate. In [5], a differentiable measure of information gain based on frontier is introduced, allowing path to be optimized with gradient information.…”