Socially assistive robots can provide assistance to seniors through social interactions at home. As these robots are not always collocated with the user, they need to first find the user before initiating an interaction. The majority of robot person search techniques in home environments do not consider dynamic user behaviours. This thesis presents a robot person search planner that considers evidence, which are objects at home that are associated with a user’s current location. By checking the presence or state of an evidence during the search, the planner can narrow down the current user location and find the person more quickly by searching fewer regions. An online Partially Observable Markov Decision Process (POMDP) solver is used to generate a sequence of actions that the robot can execute until the user is found. The planner outperformed techniques that do not consider such objects in both cases where the user is static and dynamic. The search planner was integrated into a mobile robot and tested in a home environment with real users.