An essential aspect of human robot interaction is proac-tive robot behavior particularly in situations where the robot is able to determine by itself if, how and when it can intervene and help. This is certainly valuable since it permits the user to be freed from the burden of permanently monitoring the robot and choosing the command that should be issued to the robot. In this work we present an architecture for proactive robot behavior. Its main features involve the ability to select high level goals based on scenario recognition. The goals are then refined by a specific planner that is able to determine if the robot can contribute to the goal achievement and finally a human aware supervision system that allows the robot to share the human activity thanks to its ability to achieve task cooperatively. The paper describes the overall system and its implementation on a realistic testbed.