Carnegie Mellon University (CMU) researchers are using the videogame "Doom" as an artificial intelligence (AI) research platform as part of the Visual Doom AI Competition (VizDoom).
The researchers' program finished second in each of two competitions against Intel and Facebook, respectively. The team spent four months working to create its autonomous AI agent to play first-person shooter games, and team members say the program's architecture substantially outperforms built-in AI agents of the game as well as humans in death-match scenarios.
The researchers say their algorithm represents a step forward in technology and eventually could enable robots and self-driving vehicles to better navigate real-world environments and perform more complex tasks. The algorithm uses "deep reinforcement learning" to reward successful actions and punish failures, enabling it to master human-level ability in various tasks, including object recognition and high-dimensional robotic control.
The researchers built the program on top of Google's DeepMind and separated the task of navigating in a three-dimensional environment, and identifying and retrieving objects from the task of encountering enemies and avoiding being shot. The researchers also incorporated short-term memory into the program, so the agent could better understand motion.
From Pittsburgh Post-Gazette
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found