Quantum Frog: Quantized-Time Cooperative Game Analyzed via RL
Researchers have unveiled Quantum Frog, a cooperative game for two players featuring a unique quantized-time mechanic that allows the environment to progress only with player actions. Drawing inspiration from Frogger, the objective involves two frogs navigating an 8×8 grid filled with traffic. The team utilizes reinforcement learning (RL) to tackle four main design challenges: scaling difficulty based on traffic density, determining the best single-agent strategy, assessing the cooperation gap between independent and cooperative gameplay, and developing emergent joint strategies. The agents undergo training using Tabular Q-Learning, Deep Q-Network (DQN), Independent DQN (IDQN), and Multi-Agent Proximal Policy Optimization (MAPPO with a centralized critic), tested against varying traffic densities ranging from one to six cars.
Key facts
- Quantum Frog is a two-player cooperative game with quantized-time mechanic.
- Game inspired by Frogger, requires two frogs to cross an 8×8 grid of traffic.
- Reinforcement learning used to analyze difficulty scaling, optimal policy, cooperation gap, and emergent strategies.
- Agents trained via Tabular Q-Learning, DQN, IDQN, and MAPPO.
- Evaluated against traffic densities of one to six cars.
- Study answers four design questions about game mechanics and agent cooperation.
Entities
—