Active SLAM Utility Function Exploiting Path Entropy

TL;DR

This work addresses efficient Active SLAM frontier selection by introducing a utility that blends path entropy with a D-Optimality criterion on the pose-graph to guide exploration. It computes path-entropy for frontier paths using occupancy-grid probabilities and augments the NP4 frontier weighting with a D-optimality-based term derived from the weighted graph Laplacian, encapsulated as $D$-Optimality $D-Opt = \exp\left( \frac{1}{n} \sum_{k=1}^n \log(\zeta_k) \right)$. Through simulations in Willow Garage and Office environments and real-world experiments with ROSBot2, the approach achieves higher graph connectivity, improved map quality (SSIM) and lower RMSE, and faster coverage than frontier-based or AGS baselines. It demonstrates scalable, uncertainty-aware exploration with potential for multi-robot frontier sharing. The method integrates path-entropy-driven exploration with a D-optimality backbone to robustly reduce map and pose uncertainty while increasing unknown-area discovery.

Abstract

In this article we present a utility function for Active SLAM (A-SLAM) which utilizes map entropy along with D-Optimality criterion metrices for weighting goal frontier candidates. We propose a utility function for frontier goal selection that exploits the occupancy grid map by utilizing the path entropy and favors unknown map locations for maximum area coverage while maintaining a low localization and mapping uncertainties. We quantify the efficiency of our method using various graph connectivity matrices and map efficiency indexes for an environment exploration task. Using simulation and experimental results against similar approaches we achieve an average of 32% more coverage using publicly available data sets.

Figures (8)

• Figure 1: Graph SLAM structure
• Figure 2: Framework and proposed utility function
• Figure 3: Proposed utility function ROS implementation, purple line = Bresenham's line, sphere = frontier candidate, Green Squares = frontier centroids.
• Figure 4: Uncertainty evolution of AGS, FD and Our in \ref{['fig3:1a']} Willow Garage, \ref{['fig3:1b']} Office Environments.
• Figure 5: Comparison of evolution of map discovered with average and standard deviation for our, AGS and FD methods. With 15 simulations (30 minutes each for every method)