Where do you believe you are going, human? Toward active intention recognition in unknown environments for joint robot-human search and rescue missions

Search and rescue missions are particularly dan- gerous tasks for first responders. Fully automating even the riskiest and earliest phases is still hindered by limited au- tonomous exploration algorithms. This work proposes a first step toward extending robot use in search and rescue missions by allowing them to anticipate human first responder actions, thus enabling exploration fo- cused on areas more relevant for the responder and thus for the task. The robot controller implements an active intention recogni- tion paradigm to perceive, even under sensory constraints, not only the target position but also the first responders movements which can inform on her intentions: reaching the position where she expects the target to be. This is implemented using an extension of MonteCarlo based planning where reward is extended with an entropy reduction bonus.