Alfredo Nantes

Soon "agents" of artificial intelligence (AI) will wander through the landscapes of three-dimensional computer games, automating the mundane and time-consuming aspects of game-testing.

Queensland University of Technology Information Technology PhD researcher Alfredo Nantes is creating intelligent tools that can "see" visual anomalies in a 3D computer game.

"Visual anomalies are things like incorrect shadowing or lighting, texturing problems and all artefacts that corrupt the realism of the game scene," Mr Nantes said.

"Modern games are so complex, with environments rivalling the complexity of high quality cinema animations, that this environment testing process is very costly, both in time and other resources.

"Currently, the testing is done manually by people, but they may get tired and overlook something or be too involved in playing the game to see everything. Therefore an automated testing approach is very beneficial to a games company."

Combining his background in computer engineering with research into AI and Computer Vision, Mr Nantes is programming a prototype debugger agent that will continually scan the images on screen for visual mistakes as the human game-tester plays the game.

"The rate of mistake detection is very good," he said.

"The next step in using this technology is to develop AI agents to move around the game like invisible characters and check the entire game environment, independent of the human play-tester.

"The agents will move autonomously, performing certain tasks and reporting back the results."

Mr Nantes said this technology would not only improve testing accuracy, but it would also allow human game-testers to focus completely on the playability of a game.

"People need to test the entertainment factors of the game," he said.

"The agents will help them to enjoy the experience better - we want people to enjoy testing the game."

Mr Nantes said the types of 3D games this technology could be used for are first-person shooters, role-play and adventure games.

Mr Nantes will present his work at a conference at Stanford University later this year.