Heterogeneous multi-agent interaction management: from theory to practice

This thesis addresses the topic of controlling, estimating and managing heterogeneous multi-agent systems, namely interacting devices with different actuation and sensing capabilities, and the difficulties involved with migrating from theory alone to practice. For the sensing point of view, great attention is placed in inter-agent bearing sensing capabilities, where the agents can only measure their relative direction vectors expressed in their own local reference frames. This arises from the simplicity of such sensors and their processing pipelines; indeed, a bearing vector can be easily extrapolated from an optical camera and minimum computer vision algorithms. From the control point of view, gradient descent methods and Nonlinear Model Predictive Control play the lead role. The former is well suited for minimizing potential cost functions and for proving stability of the considered systems while the latter perfectly fits in the heterogeneous frameworks, with agents having different actuation constraints. A wide range of scenarios is presented. First of all, the control and stabilization of an heterogeneous bearing-based formation by the use of Bearing Rigidity Theory is investigated and the superiority of an heterogeneous control approach over a mixture of homogeneous ones is showed. Then, the localization of an uncooperative target by a group of seekers from their perturbed bearing measurements is tackled adopting an active-sense approach which maximizes the estimation accuracy. A novel NMPC trajectory controller for tilting quadrotors i proposed to fill the gap between under-actuated coplanar aerial platforms and inefficient tilted multirotors. Finally, the implementation challenges of multi agent controllers are investigated by dealing with the autonomous landing of a drone on a moving platform.