Urban Search And Rescue (USAR) is a time critical task since all survivors have to be rescued within the first 72 hours. One goal in Rescue Robotics is to support emergency response by mixed-initiative teams consisting of humans and robots. Their task is to explore the disaster area rapidly while reporting victim locations and hazardous areas to a central station, which then can be utilized for planning rescue missions. To fulfill this task efficiently, humans and robots have to map disaster areas jointly while co- ordinating their search at the same time. Additionally, robots have to perform subproblems, such as victim detection and navigation, autonomously. In disaster areas these problems are extraordinarily challenging due to the unstructured environment and rough terrain. Furthermore, when communication fails, methods that are deployed under such conditions have to be decentralized, i.e. operational without a central station. In this thesis a unified approach joining human and robot resources for solving these problems is contributed. Following the vision of combined multi-robot and multi-human teamwork, core problems, such as position tracking on rough terrain, mapping by mixed teams, and decentralized team coordination with limited radio communication, are directly addressed. More specific, RFID-SLAM, a novel method for robust and efficient loop closure in large-scale environments that utilizes RFID technology for data association, is contributed. The method is capable of jointly improving multiple maps from humans and robots in a centralized and decentralized manner without requiring team members to perform loops on their routes. Thereby positions of humans are tracked by PDR (Pedestrian Dead Reckoning), and robot positions by slippage- sensitive odometry, respectively. The joint-graph emerging from these trajectories serves as an input for an iterative map optimization procedure. The introduced map representation is further utilized for solving the centralized and decentralized coordination of large rescue teams. On the one hand, a deliberate method for combined task assignment and multi-agent path planning, and on the other hand, a local search method using the memory of RFIDs for coordination, are proposed. For autonomous robot navigation on rough terrain and real-time victim detection in disaster areas an efficient method for elevation map building and a novel approach to genetic MRF (Markov Random Field) model optimization are contributed. Finally, a human in the loop architecture is presented that integrates data collected by first responders into a multi-agent system via wearable computing. In this context, the support and coordination of disaster mitigation in large-scale environments from a central-command-post-perspective are described. Methods introduced in this thesis were extensively evaluated in outdoor environments and official USAR testing arenas designed by the National Institute of Standards and Technology (NIST). Furthermore, they were an integral part of systems that won in total more than 10 times the first prize at international competitions, such as the RoboCup world championships.