No abstract

Abstract-In this paper, we investigate the possibility of human physical activity recognition in a robot game scenario. Being able to recognize types of activity is essential to enable robot behavior adaptation to support player engagement. Also, the introduction of this recognition system will allow for development of better models for prediction, planning and problem solving in PIRGs that can foster human-robot interaction. The experiments reported on this paper were performed on data collected from real in-game activity, where a human player faces a mobile robot. We use a custom single tri-axial accelerometer module attached to the player's chest in order to capture motion information. The main characteristic of our approach is the extraction of features from patterns found on the motion variance rather than on raw data. Furthermore, we allow for the recognition of unconstrained motion given that we do not ask the players to perform target activities before hand: all detectable activities are derived from the free player motion during the game itself. To the best of our knowledge, this is the first paper to consider activity recognition in a physical interactive robogame.

Physically-Interactive RoboGames (PIRG) are an emerging application whose aim is to develop robotic agents able to interact and engage humans in a game situation. In this framework, learning a model of players' activity is relevant both to understand their engagement, as well as to understand specific strategies they adopted, which in turn can foster game adaptation. Following such directions and given the lack of quantitative methods for player modeling in PIRG, we propose a methodology for representing players as a mixture of existing player's types uncovered from data. This is done by dealing both with the intrinsic uncertainty associated with the setting and with the agent necessity to act in real time to support the game interaction. Our methodology first focuses on encoding time series data generated from player-robot interaction into images, in particular Gramian angular field images, to represent continuous data. To these, we apply latent Dirichlet allocation to summarize the player's motion style as a probabilistic mixture of different styles discovered from data. This approach has been tested in a dataset collected from a real, physical robot game, where activity patterns are extracted by using a custom three-axis accelerometer sensor module. The obtained results suggest that the proposed system is able to provide a robust description for the player interaction.