Training an Agent for Third-person Shooter Game Using Unity ML-Agents

Abstract

The development of deep reinforcement learning has been widely used in the field of games. FPS and TPS are very suitable. At the same time, there are many training platforms and simulation environments. In this paper, a third-person shooting environment is built under the environment of Unity ML-Agents and some agents are trained by PPO and actor-critic algorithm. Through this visual in-depth reinforcement learning, agents learn skills such as searching for enemies in unfamiliar areas, searching for cartridges after injury, supplying ammunition and multi-batch of dangerous targets. Experiments show that deep reinforcement learning can fully improve the intelligence level of agent in the field of game, which is obviously superior to the ordinary intelligent behavior based on state machine.