Localization and Energy-aware path planning of mobile autonomous robots

Nowadays, robots are being used in a broad number of applications. A team of robots is an effective solution for all those cases in which the primary goal is not achievable by a single unit. For instance, applications such as search-and-rescue and agriculture can benefit from a team for increasing the rate of coverage or allowing differentiating the capabilities of a single unit by carrying different sensors. The objective of this thesis is to provide instruments necessary for developing such team-based applications. In particular, it focuses on localization and energy-aware path planning techniques. In the first part of this work, the problem of localizing a group of nodes in indoor environments will be addressed. In this context, where GPS is not a viable solution, it is possible to locate a group of nodes by measuring inter-nodes distances with radio transceivers and thus deriving their coordinates. However, this is not an easy task for two main reasons. First, distance measurements suffer from the presence of obstacles that alter the actual measured value. Whatever the sensor used, objects around the nodes affect the power of the message transmitted and create the so-called multi-path phenomenon, where multiple copies of the same signal are being reflected and arrive at the receiver. Errors in the distance measurements lead to wrong estimates of the coordinates. Secondly, even without errors, finding the coordinates from relative distances is not trivial, and there exist cases in which a unique solution of the problem does not exist. Distance measurement errors then exacerbate the problem. In this work, both these issues are addressed, separately and in conjunction. Multiple sensor measurements are fused to increase the accuracy of distance estimation. A modified version of the Multidimensional Scaling (MDS), called enhanced MDS (eMDS) is proposed for reducing the ambiguities that come when the robot positions are measured with respect to a relative reference system. A novel algorithm to solve both problems is proposed, in which Non-line-of-sight measurements errors are modeled with a Gaussian mixture distribution. The parameters of such distributions are estimated and used to improve the accuracy of the coordinates that will act as a feedback for further refining the estimated distributions. The second part of the thesis focuses on energy-aware path finding for unmanned aerial vehicles (UAV). In particular, many applications that involve the use of flying vehicles suffer from the problem of correctly estimating the energy consumed by the robot during their operations. Some path-planning algorithms could benefit from the knowledge of the energy consumption of the vehicle to adapt and produce energy-aware trajectories that are i) safe to fly, b) cost-effective regarding energy, c) and prone to dynamic changes due to disturbances such as high wind. In this context, an energy model for a quadrotor is designed. The energy cost of all the possible trajectories (acceleration, constant speed, landing, etc.) is characterized by using actual measurements. Then this model is used to provide safety guarantees both offline (estimating the expected energy required for a pre-selected path) and online (checking whether if the remaining energy is enough to perform a safe return-to-home). The effects of the wind on the energy model have been studied and a run-time re-planning algorithm that changes the trajectory of the vehicle according to the actual disturbance is proposed.