Unmanned ground vehicles (UGVs) and unmanned aerial vehicles (UAVs) are promising assets to support rescue operations in natural or man-made disasters. Most UGVs and UAVs deployed in the field today depend on human operators and reliable network connections to the vehicles. However, network connections in challenged environments are often lost, thus control can no longer be exercised. In this paper, we present a novel approach to emergency communication where semi-autonomous UGVs and UAVs cooperate with humans to dynamically form communication islands and establish communication bridges between these islands. Humans typically form an island with their mobile devices if they are in physical proximity; UGVs and UAVs extend an island's range by carrying data to a neighboring island. The proposed approach uses delay/disruptiontolerant networking for non-time critical tasks and direct mesh connections for prioritized tasks that require real-time feedback. The developed communication platform runs on rescue robots, commodity mobile devices, and various drones, and supports our operations and control center software for disaster management.

Abstract-Monitoring in large scale environments is a typical mission in cooperative robotics. This task requires the exploration of a huge domain by a generally small number of sensor equipped mobile robots. As time restrictions prohibit an exhaustive global search, a sampling strategy is required that allows an efficient spatial mapping of the environment. This paper proposes an adaptive sampling strategy for efficient simultaneous tracking of multiple concentration levels of an atmospheric plume by a team of cooperating unmanned aerial vehicles (UAVs). The approach combines uncertainty and correlation-based concentration estimates to generate sampling points based on already gathered data. The adaptive generation of sampling locations is coupled to a distributed modelpredictive controller for planning optimal vehicle trajectories under collision and communication constraints. Simulation results demonstrate that connectivity of all involved vehicles can be maintained and an accurate reconstruction of the plume is obtained efficiently.

Efficient online state estimation of dynamic dispersion processes plays an important role in a variety of safety-critical applications. The use of mobile sensor platforms is increasingly considered in this context, but implies the generation of situation-dependent vehicle trajectories providing high information gain in real-time. In this paper, a new adaptive observation strategy is presented combining state estimation based on partial differential equation models of the dispersion process with a model-predictive control approach for multiple cooperating mobile sensors. In a repeating sequential procedure, based on the Ensemble Transform Kalman Filter, the uncertainty of the current estimate is determined and used to find valuable measurement locations. Those serve as target points for the controller providing optimal trajectories subject to the vehicles' motion dynamics and cooperation constraints. First promising results regarding accuracy and efficiency were obtained.

No abstract

Abstract. Optimal coordination of multiple sensors is crucial for efficient atmospheric dispersion estimation. The proposed approach adaptively provides optimized trajectories with respect to sensor cooperation and uncertainty reduction of the process estimate. To avoid the timeconsuming solution of a complex optimal control problem, estimation and vehicle control are considered separate problems linked in a sequential procedure. Based on a partial differential equation model, the Ensemble Transform Kalman Filter is applied for data assimilation and generation of observation targets offering maximum information gain. A centralized model-predictive vehicle controller simultaneously provides optimal target allocation and collision-free path planning. Extending previous work, continuous measuring is assumed, which attaches more significance to the course of the trajectories. Local attraction points are introduced to draw the sensors to regions of high uncertainty. Moreover, improved target updates increase the sampling efficiency. A simulated test case illustrates the approach in comparison to non-attracted trajectories.