Communications of the ACM
Ubisoft Uses AI to Teach a Car to Drive Itself in Racing Game
Game developer Ubisoft has tested a reinforcement-learning algorithm that can manage discrete, continuous action in a principled and predictable manner on a commercial game.
Credit: Ubisoft
Ubisoft tested a reinforcement-learning algorithm that can manage discrete, continuous video game actions in a principled and predictable manner on a commercial game.
Ubisoft developers based the algorithm on the University of California, Berkeley's Soft Actor-Critic architecture, which can learn to generalize to previously unseen conditions; the researchers extended the framework to a hybrid environment with both continuous and discrete actions.
The researchers assessed the algorithm on three settings to benchmark reinforcement-learning systems.
In a separate test, the algorithm trained a video game vehicle with two continuous actions—acceleration and steering—and one binary discrete action—hand brake—with the goal of following a given path at maximum speed in unfamiliar environments.
The researchers said their strategy can cover various approaches for a software agent to engage with a game environment, such as when the agent has the same inputs as a player.
From VentureBeat
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA
No entries found