Multi-Robot Autonomous Exploration and Mapping Under Localization Uncertainty with Expectation-Maximization

TL;DR

An autonomous exploration algorithm designed for decentralized multi-robot teams, which takes into account map and localization uncertainties of range-sensing mobile robots, and an iterative expectation-maximization inspired algorithm to assess the potential out-comes of both a local robot’s and its neighbors’ next-step actions.

Abstract

We propose an autonomous exploration algorithm designed for decentralized multi-robot teams, which takes into account map and localization uncertainties of range-sensing mobile robots. Virtual landmarks are used to quantify the combined impact of process noise and sensor noise on map uncertainty. Additionally, we employ an iterative expectation-maximization inspired algorithm to assess the potential outcomes of both a local robot's and its neighbors' next-step actions. To evaluate the effectiveness of our framework, we conduct a comparative analysis with state-of-the-art algorithms. The results of our experiments show the proposed algorithm's capacity to strike a balance between curbing map uncertainty and achieving efficient task allocation among robots.

Figures (7)

• Figure 1: Problem setup. In this $100 m \times 80 m$ virtual map created by a two-robot team, gray ellipses depict the uncertainty of visited cells. The green robot's position is denoted by a black star, and its newly selected target state is represented by a red star. The current position of the other robot on the team is marked as a black rectangle. Landmarks are expressed by black x's. Potential frontiers emerge along the boundary between the explored and unexplored areas. The three types of frontiers—exploring, revisiting, and rendezvous—are denoted by their respective colors: green, blue, and purple.
• Figure 2: System Architecture. The pipeline of the proposed approach.
• Figure 3: EM-based uncertainty propagation with virtual observations. Nodes representing the current robot states are distinguished by red edges. For every potential next target state (shown in pink), a trajectory simulation is executed, leading to the generation of a sequence of virtual observations indicated by dashed arrows. Simultaneously, we model the future states and observations of other robots as they approach their individual current target states.
• Figure 4: 50 trials of three robots navigating in 100m x 100m environments with 20 landmarks, each with a radius of 1m. They are exploring the environment by constructing a virtual map with cell size of $2 m$. At left, the average robot localization error for each robot state, at center, the average landmark position error for each landmark, and at right, the explored ratio, all plotted against distance.
• Figure 5: 50 trials of three robots navigating in 200m x 200m environments with 20 landmarks, each with a radius of 1m. They are exploring the environment by constructing a virtual map with cell size of $4 m$. At left, the average robot localization error for each robot state, at center, the average landmark position error for each landmark, and at right, the explored ratio, all plotted against distance.