We present the architecture of a mobile robot whose goal is to perform autonomous semantic mapping of indoor environments. This robot has been developed under the PACOM project in order to participate in the "CAROTTE" challenge organised by the french Armament Procurement Agency (DGA) and Research Funding Agency (ANR). This challenge takes place in an arena of approximately 100 m2 wherein objects and obstacles are placed, which contains several rooms, with variable ground types and difficulties (fitted carpet, tiling, grid, gravel, ...). Several types of objects are present in multiple instances which must be detected, located, and identified by the robot.