In this project, a general research and education
platform for artificial intelligence (AI) called OpenNERO will be built. The
platform is based on a sophisticated simulation and graphical display
of a 3-D physical world, including multiple agents with embedded
sensors and effectors, multiple objects, together with GUIs for
manipulating the environment and the agents. Using the existing NERO
game as a starting points, our goals are to (1) implement the game
entirely with open source software, (2) extend it to more general
physical and communicative interactions among the agents and the
agents and the environment, (3) build tools for the user to construct
new environments, tasks, and AI methods, and to visualize the
behaviors and analyze performance statistically, and (4) implement and
demonstrate a number of AI techniques, such as search, vision, and
supervised and reinforcement learning. The resulting OpenNERO
software environment will allow developing and testing new AI methods
as well as demonstrating existing methods in a sophisticated and
concrete simulation of the physical world. It will therefore serve as
a catalyst for research in intelligent agents, as well as a
demonstration software for AI education.
This research is supported by Google, Inc. under its Faculty Research Awards program.