An Intelligent Human Simulation (InHuS) for developing and experimenting human-aware and interactive robot abilities
Résumé
Testing and experimenting software which synthesize human-aware and interactive robot abilities is a challenging task. To answer the lack of individual intelligent agents in simulation, this project aims to develop a generic autonomous simulated human both reactive and rational specially designed to act and interact in a simulated environment where a robot acts and interacts. A generic architecture called InHuS has been designed for the deployment of such a system. A first implementation has then been implemented in order to investigate more deeply its pertinence and usability. The chosen context is social navigation since our team has contribution in this specific topic and also because there exist today a number of available human-aware robot navigation algorithms. Several experiments have been conducted with different robot navigation planners showing that the system is able to create pertinent situations and also long-term runs situations for testing, challenging and measuring the performance of such planners. These experiments also allowed to exhibit a set of tools and metrics for the analysis of robot behavior over time. Hence, we claim that InHuS is a pertinent approach to provide a suitable framework for the first steps of debugging, experimenting and tuning human-aware and interactive robot abilities before their final evaluation with real naive users.
Domaines
Robotique [cs.RO]
Origine : Fichiers produits par l'(les) auteur(s)