MAgent: A Many-Agent Reinforcement Learning Platform for Artificial Collective Intelligence

By Lianmin Zheng, Jiacheng Yang, Han Cai, Weinan Zhang, J. Wang, Yong Yu. 2017

248 citations recorded by the source corpus.

We introduce MAgent, a platform to support research and development of many-agent reinforcement learning. Unlike previous research platforms on single or multi-agent reinforcement learning, MAgent focuses on supporting the tasks and the applications that require hundreds to millions of agents. Within the interactions among a population of agents, it enables not only the study of learning algorithms for agents' optimal polices, but more importantly, the observation and understanding of individual agent's behaviors and social phenomena emerging from the AI society, including communication languages, leaderships, altruism. MAgent is highly scalable and can host up to one million agents on a single GPU server. MAgent also provides flexible configurations for AI researchers to design their customized environments and agents. In this demo, we present three environments designed on MAgent and show emerged collective intelligence by learning from scratch.

Subfields: Artificial Intelligence, Reinforcement Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Reinforcement Learning

Citations in the corpus

Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning (2020, background): “…al., 2017), Keepaway Soccer (Stone et al., 2005), or simple gridworld-like environments (Lowe et al., 2017; Leibo et al., 2017; Yang et al., 2018; Zheng et al., 2017), there are currently no challenging standard testbeds for centralised training with decentralised execution with the exception of…”
The StarCraft Multi-Agent Challenge (2019, background): “Yang et al. (2018); Zheng et al. (2017) present a framework for creating gridworlds focuses on many-agents tasks, where the number of agents ranges from the hundreds to the millions.”
From Few to More: Large-scale Dynamic Multiagent Curriculum Learning (2019, methodology): “In this section, we evaluate the performance of our DyMA-CL on two large-scale scenarios: 1) StarCraft II, which contains various scenarios for a number of agents to learn coordination to solve complex tasks; and 2) MAgent (Zheng et al. 2018), which is a simulated battlefield with two large-scale ar”
Factorized Q-learning for large-scale multi-agent systems (2018, background): “The recently emerged open source multi-agent reinforcement learning platform MAgent (Zheng et al. 2018) enable us to simulate battles between two armies (groups) in which the soldiers from one army would cooperate with each other to ﬁght against their enemies, i.e., the soldiers from the other army.”
Mean Field Multi-Agent Reinforcement Learning (2018, methodology): “The Battle game in the Open-source MAgent system (Zheng et al., 2018) is a Mixed CooperativeCompetitive scenario with two armies fighting against each other in a grid world, each empowered by a different RL algorithm.”
Reinforcement Learning With Task Decomposition for Cooperative Multiagent Systems (2020, background): “It will bring more difficulties when the connection topology between agents is complex [5]–[11].”
Influencing Long-Term Behavior in Multiagent Reinforcement Learning (2022, unknown): “How does FURTHER scale to a large number of agents? Finally, we show the scalability of our method regarding the number of agents using the battle domain ([32]; see Figure 3e).”
PettingZoo: Gym for Multi-Agent Reinforcement Learning (2020, unknown): “This design choice has become the standard for MARL outside of strictly turn-based games like poker, where simultaneous stepping would be a poor conceptual ﬁt [Lowe et al., 2017, Zheng et al., 2017, Gupta et al., 2017, Liu et al., 2019, Liang et al., 2018, Weng et al., 2020].”
Efficient Large-Scale Fleet Management via Multi-Agent Deep Reinforcement Learning (2018, background): “A straightforward extension of independent DQN from small scale to a large number of agents, is to share network parameters and distinguish agents with their IDs [30].”
A Speculative Study on 6G (2019, background): “Collective AI is a related concept that has emerged recently to deal with the situation where multiple AI agents aim to achieve the same goal based on local training at each AI agent, with limited or no direct communication amongst the agents [10].”
Neural MMO: A Massively Multiagent Game Environment for Training and Evaluating Intelligent Agents (2019, unknown): “Most such works further focus on learning a speciﬁc dynamic, such as predator-prey Yang et al. (2018b) or are more concerned with the study than the learning of behavior, and use hard-coded rewards Zheng et al. (2017).”
Graph Convolutional Reinforcement Learning (2018, methodology): “For the experiments, we adopt a grid-world platform MAgent (Zheng et al., 2017).”