Embodied Artificial Intelligence (Embodied AI) emphasizes agents' ability to perceive, understand, and act in physical environments. Simulation platforms play a crucial role in advancing this field by enabling the validation and optimization of algorithms. However, existing platforms face challenges such as multilevel technical integration complexity, insufficient modularity, interface heterogeneity, and adaptation to diverse hardware. We present BestMan, a simulation platform based on PyBullet, designed to address these issues. BestMan introduces an integrated multilevel skill chain for seamless coordination across perception, planning, and control; a highly modular architecture for flexible algorithm integration; unified interfaces for smooth simulation-to-reality transfer; and a hardware-agnostic approach for adapting to various mobile manipulator configurations. These features collectively simplify development and enhance platform expandability, making BestMan a valuable tool for Embodied AI research.

翻译：具身人工智能（Embodied AI）强调智能体在物理环境中感知、理解与行动的能力。仿真平台通过支持算法的验证与优化，在推动该领域发展中扮演着关键角色。然而，现有平台面临诸多挑战，例如多层次技术集成的复杂性、模块化程度不足、接口异构性以及对多样化硬件的适配困难。本文提出BestMan，一个基于PyBullet的仿真平台，旨在解决上述问题。BestMan引入了集成的多层次技能链，以实现感知、规划与控制间的无缝协同；采用高度模块化的架构，以支持灵活的算法集成；提供统一的接口，以实现平滑的仿真到现实迁移；并采用硬件无关的方法，以适应各种移动机械臂配置。这些特性共同简化了开发流程并增强了平台的可扩展性，使BestMan成为具身人工智能研究的一个有力工具。