Vectorized Multi-Agent Simulator (VMAS)

该数据集是一个开源的二维物理模拟平台，旨在评估多种多智能体强化学习（MARL）算法在不同多智能体协同问题上的表现。该平台提供了多样化的场景，这些场景需要不同程度的个体技能和协作，且动作空间可以是连续的或离散的。其规模覆盖了各种多智能体协同场景，任务主要集中在基于低维向量观测的多智能体协同问题。

This dataset is an open-source 2D physics simulation platform designed to evaluate a variety of multi-agent reinforcement learning (MARL) algorithms across different multi-agent collaborative tasks. The platform provides diverse scenarios that require varying degrees of individual skill and collaboration, with action spaces that can be either continuous or discrete. It covers a wide range of multi-agent collaborative scenarios, with tasks primarily focusing on multi-agent collaborative problems based on low-dimensional vector observations.

搜集汇总

背景与挑战

背景概述

VMAS是一个专为多智能体强化学习设计的向量化可微分模拟器，包含一个基于PyTorch的2D物理引擎和多种挑战性场景。它支持弹性碰撞、关节、自定义重力等特性，并与多种强化学习框架兼容，适用于高效的MARL基准测试和研究。