Dezentrale kooperative Exploration von Umweltfeldern mit kleinen autonomen Unterwasserrobotern in räumlich beschränkten Umgebungen

Exploration and monitoring of environmental and potential hazardous fields in underwater environments, such as chemical concentration or radiation, are among the most promising tasks to be performed by groups of autonomous underwater vehicles (AUVs). While the first phase of this project concentrated on teams of AUVs in structured and fully known environments, e.g. an obstacle-free water basin, the second phase extends this work to confined and cluttered scenarios. To date, little work has examined the challenges that arise when moving from operation in controlled research environments to field exploration in real-world cluttered scenarios. Examples range from harbor basins to industrial facilities such as legacy nuclear decommissioning ponds. The main objective of this research project is the decentralized exploration and monitoring of environmental fields with teams of small-scale underwater robots. Thereby, the team of micro-AUVs coordinates their motion to estimate an unknown concentration field within the fluid volume of interest. The environmental field is modeled using a stochastic field belief representation, namely non-stationary Gaussian Markov Random Fields (GMRFs). Within this setting, robot agents acting as mobile sensor nodes compute information-rich paths that reduce the uncertainty of the estimated field effectively. Due to the massive limitations with respect to communication bandwidth in the underwater domain, approaches using a central coordinating unit render infeasible. As a consequence, decentralized concepts are investigated that rely on computationally lightweight field estimation and motion planning methods that allow implementation onboard the individual micro underwater robots. In these cluttered environments, obstacles such as walls induce anomalies in the field structure that are a major challenge and need to be included within the field belief and exploration framework. Since the unknown fields are modeled by a nonparametric approach (GMRFs) a minimum of a priori knowledge is required. The initially unknown covariance function can be estimated along the field. The approach requires analyzing a fundamental tradeoff in explorative problems summarized by the following question: Shall future observations be collected to improve the field model (a better covariance function estimate), or shall observations be taken at locations where the current (possibly inaccurate) model indicates the highest uncertainty? In the first phase of the project, a computationally lightweight field belief representation, based on GMRFs has been derived. Two planning and control strategies for autonomous exploration of environmental fields have been developed. These preliminary results are the starting point for extending a truly decentralized exploration framework in confined and cluttered environments. A thorough analysis and comparison of these strategies will be performed, and the concept will be verified experimentally.

Options