Decentralized multi-agent exploration with online-learning of Gaussian processes

Abstract

Exploration is a crucial problem in safety of life applications, such as search and rescue missions. Gaussian processes constitute an interesting underlying data model that leverages the spatial correlations of the process to be explored to reduce the required sampling of data. Furthermore, multiagent approaches offer well known advantages for exploration. Previous decentralized multi-agent exploration algorithms that use Gaussian processes as underlying data model, have only been validated through simulations. However, the implementation of an exploration algorithm brings difficulties that were not tackle yet. In this work, we propose an exploration algorithm that deals with the following challenges: (i) which information to transmit to achieve multi-agent coordination; (ii) how to implement a lightweight collision avoidance; (iii) how to learn the data's model without prior information. We validate our algorithm with two experiments employing real robots. First, we explore the magnetic field intensity with a ground-based robot. Second, two quadcopters equipped with an ultrasound sensor explore a terrain profile. We show that our algorithm outperforms a meander and a random trajectory, as well as we are able to learn the data's model online while exploring.

Distributed Multi-Robot Information Gathering under Spatio-Temporal Inter-Robot Constraints

“…In contrast to [6,13], here we argue for the use of a decentralized architecture to decide robots movement. Multiple works proposed decentralized IG methods with GPs; see, e.g., [1,4,13,14]. The previous techniques [1,4,13,14] employ a discretization of the robot state and action spaces and search-based planning algorithms for exploration.…”

“…Multiple works proposed decentralized IG methods with GPs; see, e.g., [1,4,13,14]. The previous techniques [1,4,13,14] employ a discretization of the robot state and action spaces and search-based planning algorithms for exploration. Thus, they cannot consider kinematic constraints associated to the robot motion for non-holonomic vehicles, like fixed-wing aircrafts.…”

“…Information gathering (IG) is a key task in many robotic applications such as, e.g., magnetic field mapping [1], environmental monitoring [2], or wind field mapping [3]. The IG task can clearly benefit from distributed multi-robot coordination strategies: First, by means of parallelization, as tasks could be split between robots, and second, in terms of robustness, as tasks of a faulty robot could be overtaken by the rest of the team.…”

“…A common approach often used in the literature to solve multi-robot information gathering (MR-IG) tasks is to use an underlying model of the physical process under study that, together with an information-theoretic metric, is employed to predict the impact of certain robot actions and states (see, e.g., [1,4]). In particular, in this work we assume Gaussian processes (GPs) for regression [5] as underlying model, and mutual information (MI) [6] as information metric.…”

“…GPs are state-of-the-art models to represent spatio-temporal fields [1][2][3]5]. Furthermore, the use of GPs together with MI has been shown to significantly outperform former MR-IG algorithms [7,8].…”