Socially assistive robotics considers robots helping people via social interactions. With technological advancements over the past few decades, there has been an increase in the number and types of socially assistive robotic applications. However, these applications have focussed on scenarios in which people begin collocated with the robot. Yet, in real-world environments, people go about their day-to-day lives and robots must first find users prior to assisting. To address this limitation, this thesis considers the multi-robot person search problem, defined herein as the person search team orienteering problem (PSTOP). The PSTOP considers dynamic users with changing locations during the search, a team of robots that may need to search a region multiple times, and a specified search time frame. User locations are modelled by novel activity probability density functions (APDFs) using past location data. The APDF is the only existing model which can determine the probability of a user occupying a region by considering when the region was previously searched. A two-stage multi-robot person search system (2-MRPSS) is introduced to generate team plans specifying when and for how long to search regions. The 2-MRPSS is the only planner which can consider user location data when determining repeated searches of a region. The 2-MRPSS is also the only existing planner which reasons about past user location data, while being both aware and strongly-coordinated. Moreover, this thesis is the first to deploy a multi-robot multi-user person search in a real-world human-centered environment by using a novel centralized architecture to oversee the generation, execution, and monitoring of the team plan. Within the architecture, this thesis introduces a novel low-cost information gathering technique and a navigation technique that adheres to human-etiquette. Simulations have been conducted to evaluate the theoretical contributions of the thesis and experiments have been conducted to evaluate the practical contributions of the thesis. The simulations show that the proposed approach finds more target users than existing state-of-the-art techniques. The experiments show that the architecture, information gathering, navigation, and 2-MRPSS can be used to successfully perform the multi-robot person search in a real-world human-centered environment.