机器人仿真到真实迁移综述、环境与基准测试1综述论文汇总按年份倒序The Reality Gap in Robotics: Challenges, Solutions, and Best Practices2025系统拆解机器人领域 “现实鸿沟” 的动力学、感知传感、执行控制三大核心成因完整覆盖从经典域随机化、域适应到前沿世界模型、可微物理的全链路解决方案同时建立了标准化的 Sim2Real 评估指标与最佳实践体系。链接https://arxiv.org/abs/2510.05247A Survey of Sim-to-Real Methods in RL: Progress, Prospects and Challenges with Foundation Models2025从强化学习 MDP 的状态、动作、转移、奖励四大核心要素构建技术分类框架首次系统梳理基础模型 / 大模型赋能的 Sim2Real 新范式覆盖机器人、自动驾驶等多领域落地场景。链接https://arxiv.org/abs/2502.11345A Survey on Sim-to-Real Transfer Methods for Robotic ManipulationSISY 2024链接https://www.researchgate.net/publication/385575540_A_Survey_on_Sim-to-Real_Transfer_Methods_for_Robotic_ManipulationSim-to-Real Transfer for Legged Robots: A Systematic Review2024聚焦足式 / 人形机器人专项系统总结动力学随机化、系统辨识、残差学习等足式机器人专属 Sim2Real 技术路线配套标准化的迁移效果评测流程。链接https://ieeexplore.ieee.org/document/10678932How simulation helps autonomous driving: A survey of sim2real, digital twins, and parallel intelligenceTIV 2023链接https://ieeexplore.ieee.org/document/10242366Parallel learning: Overview and perspective for computational learning across Syn2Real and Sim2RealIEEE/CAA Journal of Automatica Sinica, 2023链接https://ieeexplore.ieee.org/document/10057176A Brief Survey of Sim2Real Methods for Robot Learning2022链接https://link.springer.com/chapter/10.1007/978-3-031-04870-8_16A survey of sim-to-real transfer techniques applied to reinforcement learning for bioinspired robots2021链接https://ieeexplore.ieee.org/document/9552429Crossing the reality gap: A survey on sim-to-real transferability of robot controllers in reinforcement learningIEEE Access, 2021链接https://ieeexplore.ieee.org/document/9606868Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a SurveyIEEE Symposium Series on Computational Intelligence, 2020链接https://ieeexplore.ieee.org/document/93084682机器人仿真环境按年份倒序NVIDIA Isaac Sim / Isaac Lab原 Isaac Gym 升级2024基于 NVIDIA Omniverse 与 PhysX 5 构建的 GPU 加速仿真平台支持万级环境并行训练RTX 光追渲染大幅缩小视觉 Sim2Real 差距是当前工业界人形机器人、操作机器人 Sim2Real 训练的主流底座。链接https://developer.nvidia.com/isaac-simGenesisMIT CSAIL2024新一代神经物理仿真引擎支持端到端可微物理仿真精度逼近真实物理世界主打零样本 Sim2Real 迁移是 2024 年具身智能仿真领域的标志性工作。链接https://genesis.csail.mit.edu/Habitat 3.02023面向导航与家居交互的高保真仿真环境支持人形机器人、移动操作机器人多任务训练配套大量真实扫描的 3D 室内场景是导航 交互类 Sim2Real 的基准平台。链接https://aihabitat.org/Gazebo Harmonic2023ROS 生态标配的开源通用机器人仿真平台支持多物理引擎与丰富传感器插件广泛用于移动机器人、工业机器人的 Sim2Real 部署验证。链接https://gazebosim.org/SAPIEN2022面向具身智能的高保真物理仿真引擎支持刚体、铰接物体、可变形物体与流体仿真配套标准化操作任务基准是视觉 - 操作类 Sim2Real 研究的核心开源平台。链接https://sapien.ucsd.edu/CALVIN: A Benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Tasks2021链接http://calvin.cs.uni-freiburg.de/iGibson 2.0斯坦福大学2021基于真实房屋扫描重建的大规模交互仿真环境支持完全可交互的家居场景与日常操作任务主打复杂长时程任务的 Sim2Real 迁移研究。链接https://svl.stanford.edu/igibson/robosuite: A Modular Simulation Framework and Benchmark for Robot Learning2020链接https://robosuite.ai/dm_control: Software and tasks for continuous control2020链接https://github.com/google-deepmind/dm_controlSoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation2020链接https://sites.google.com/view/softgymAssistive Gym: A Physics Simulation Framework for Assistive RoboticsICRA 2020链接https://github.com/Healthcare-Robotics/assistive-gymMeta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement LearningCoRL 2019链接https://github.com/rlworkgroup/garageContinuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments2017链接https://arxiv.org/abs/1710.03641PyBullet: Real-Time Physics Simulation2017链接https://pybullet.org/wordpress/OpenAI Gym2016链接https://gym.openai.com/MuJoCo: A physics engine for model-based control2012链接https://ieeexplore.ieee.org/document/63861093机器人 Benchmarks按年份倒序REALM: A Real-to-Sim Validated Benchmark for Generalization in Robotic Manipulation2025经过真实世界严格验证的操作泛化基准系统设计视觉、语义、动力学三类扰动量化评测策略在 Sim2Real 迁移中的鲁棒性仿真与真实性能高度对齐。链接https://realm-benchmark.github.io/RoboTwin Dual-Arm Collaboration BenchmarkCVPR MEIS Workshop, 2025双臂协作操作专项基准配套仿真与真实硬件平台包含 17 项刚性 / 可变形 / 触觉类任务是双臂机器人 Sim2Real 迁移效果的标准化评测赛道。链接https://robotwin-benchmark.github.io/Robust Gymnasium: A Unified Modular Benchmark for Robust Reinforcement Learning2024链接https://github.com/SafeRL-Lab/Robust-GymnasiumHumanoid-Gym: Reinforcement Learning for Humanoid Robot with Zero-Shot Sim2Real Transfer2024链接https://sites.google.com/view/humanoid-gym/DISCOVERSE: Efficient Robot Simulation in Complex High-Fidelity Environments2024链接https://air-discoverse.github.io/NeuronsGym: A Hybrid Framework and Benchmark for Robot Tasks with Sim2Real Policy Learning2024链接https://ieeexplore.ieee.org/document/1389727ManiSkill2: A Unified Benchmark for Generalizable Manipulation Skills2024基于 SAPIEN 构建的通用机器人操作基准覆盖 20 操作任务、上千种物体支持刚体 / 可变形物体操作是当前操作类 Sim2Real 迁移能力的核心评测基准。链接https://maniskill2.github.io/BEHAVIOR-1K: A Benchmark for Embodied AI with 1000 Everyday Activities2024斯坦福大学推出的日常家居任务基准覆盖 1000 真实家庭活动任务基于 iGibson 高保真仿真主打复杂长时程任务的 Sim2Real 迁移评测。链接https://behavior.stanford.edu/RRLS : Robust Reinforcement Learning Suite2023链接https://github.com/SuReLI/RRLSLIBERO: A Benchmark for Lifelong Robot Learning2023语言条件长序列操作任务基准包含 4 大任务套件、130 细分任务专门评测策略的跨场景泛化与 Sim2Real 迁移能力是语言 - 机器人方向的标准评测集。链接https://libero-project.github.io/ManipulaTHOR: A Framework for Visual Object ManipulationCVPR 2021链接https://ai2thor.allenai.org/manipulathor/safe-control-gym: A Unified Benchmark for Safe Reinforcement Learning2021面向无人机、足式机器人的安全控制基准提供仿真与真实硬件的标准化评测流程是控制类 Sim2Real 迁移的经典评测平台。链接https://github.com/utiasDSL/safe-control-gymRLBench: The Robot Learning Benchmark Learning Environment2019链接https://sites.google.com/view/rlbench