We consider scenarios from the real-time strategy game StarCraft as new for reinforcement learning algorithms. We propose micromanagement, which present the problem of the short-term, low-level control of army during a battle. From a reinforcement learning point of view, these are challenging because the state-action space is very large, and there is no obvious feature representation for the state-action function. We describe our approach to tackle the micromanagement with deep neural network controllers from raw state features given by game engine. In addition, we present a heuristic reinforcement learning which combines direct exploration in the policy space and. This algorithm allows for the collection of traces for using deterministic policies, which appears much more efficient than, example, {\ε}-greedy exploration. Experiments show that with this, we successfully learn non-trivial strategies for scenarios with of up to 15 agents, where both Q-learning and REINFORCE struggle.
No takes yet. Share an insight, caveat, or question.
Usunier et al. (2016) studied this question.