Intelligent control and coordination of an Au-tonomous Mobile Robot (AMR) fleet is a challenging problem. Optimization of the fleet performance using centralized control is feasible only for small fleets. Decentralized control, on the other hand, is feasible also for large fleets and features flexibility and robustness. We emphasize this flexibility and robustness and consider intelligent control as the capability of a system of mobile robots to manage and handle as many maintenance tasks as possible at any time. We focus on task assignments and use entropy in a dualistic way. For task assignments, maximization of potential entropy is sought for the set of tasks supported by available AMRs. By contrast, task performance is considered by minimization of the expected actual entropy of the AMR's task execution. Distributed fleet control is proposed with a contract negotiation protocol utilizing the entropy maximization and minimization measures.

Maintenance tasks represent a potential area for applying multi-purpose Autonomous Mobile Robots (AMRs). Intelligent control and coordination of such a system is challenging and optimization methods are feasible only for small fleets. Decentralized control can provide flexibility and robustness, which are better applicable also for large fleets, though with less guaranteed performance. Our focus is on flexibility and robustness in task scheduling and task assignments and we use entropy as an indirect performance criterion for coordination, both at the system level (maximize entropy) and at the AMR level (minimize entropy). As a distributed coordination scheme, we use a modified contract negotiation protocol. We show preliminarily the feasibility of our approach with simulation results.

This paper presents the development and testing of end-to-end Neural Network (NN) controllers for automated pile loading with a robotic wheel loader. NNs were trained using the Learning from Demonstration approach, i.e. by first recording sensor and control signals during manually-driven pile loading actions. Training made use of three input signals: boom angle, bucket angle and hydrostatic driving pressure; and three output signals: boom control, bucket control and the gas command. Most testing was conducted using NNs with 5 neurons in a single hidden layer, which were able to fill the bucket reasonably well. Qualitative comparisons were made to ascertain how the amount of training data and number of hidden neurons affects bucket filling performance, for NNs trained using both the Levenberg-Marquardt and Bayesian Regularization backpropagation algorithms. Different NNs trained with the same data were also compared. An additional pile transfer experiment compared the performance of an NN controller with a heuristic automated controller and manual human control. By estimating the total volume of material transferred using 3D laser scans, human control was found to have the highest performance, though the NN outperformed the heuristic controller. This indicated that end-to-end NN control trained by demonstration could offer improvement over current heuristic methods for automated pile loading.