Planning in Inhabited Environments Human-Aware Task Planning and Activity Recognition

Our work addresses issues related to the cohabitation of service robots and people in unstructured environments. We propose new planning techniques to empower robot means-end reasoning with the capability of taking into account human intentions and preferences. We also address the problem of human activity recognition in instrumented environments. We employ a constraint-based approach to realize a continuous inference process to attach a meaning to sensor traces as detected by sensors distributed in the environment. 1 The Dream of a Robot Butler State-of-the-art commercial robots are finally finding their way into our homes (as the Roomba vacuum cleaners from iRobot) and into our workplaces (e.g., the Magic Shelf from KIVA systems), performing efficiently their tasks. However, we are still far from fulfilling the expectations raised by decades of science fiction. One of the main features that distinguished fictional robots and intelligent environments from the ones we can now deploy in our homes is the capability to closely cooperate with humans. This cooperation is both explicit and implicit, as science fiction intelligent systems can act when directly prompted by the users, but they can also support them by understanding what are their intentions and performing high level reasoning to achieve complex tasks to assist them. Human-robot cooperation and interaction is a complex research problem, not limited to the identification of suitable communication interfaces. The presence of M. Cirillo AASS Research Centre, Örebro University, Sweden Tel: +46-19-303694 E-mail: [email protected] humans in the space where robots operate has also a profound influence on how the embodied agents should perform high level reasoning and plan their actions. We believe that the interaction and teamwork of robots and humans can be greatly improved if the robot can anticipate the forthcoming actions of the human. The robots should acquire knowledge about people’s current and predicted future activities and use such knowledge to plan their actions accordingly. In the Ph.D. thesis “Planning in Inhabited Environments: Human-Aware Task Planning and Activity Recognition” [1], we propose techniques for human activity recognition in intelligent environments and humanaware robot task planning as means to achieve a better integration between people and service robots. 2 A Motivating Scenario Malin lives alone in her small apartment. The apartment has been instrumented with a series of sensors and actuators which help her manage some of the physical and cognitive difficulties she has due to age. The home alerts her if she appears to be over-cooking her meals, can recognize when she is sleeping, eating and performing other usual activities at home, and can be easily set up to monitor and respond to the occurrence of specific patterns of behavior, like getting her a drink from the fridge when she watches TV. A robotic vacuum cleaner is also deployed in the apartment. It can plan its cleaning tasks accord1 The PDF version of the thesis can be downloaded from http://oru.diva-portal.org/smash/get/diva2: