In this paper we construct an office-use autonomous mobile robot which predicts the state (either stressed, relaxed, usual, or non-existent) of a person at different places and navigates between the places. The productivity of an office worker in advanced countries is a crucial concern and we believe autonomous mobile robots without network connection and with a privacy switch are preferred to privacy-offending solutions such as monitoring cameras. We exploit recent advances in hardware and software to keep the construction cost low. The state prediction of a person in an office is based on support vector machines with image processing with stereo vision. The robot navigation between the prediction places is based on a look-around solution that we devised. Experiments using 8 hours of data gave promising results.