bulletphysics/bullet3
Classic open-source physics simulation engine (Python interface to Bullet Physics); a standard entry point for robot modeling, collision and dynamics, and the default simulation environment in many papers and courses.
google-deepmind/mujoco
High-performance contact-dynamics physics engine (open-sourced free by DeepMind in 2022); a de facto standard for legged and dexterous-hand RL and VLA research.
isaac-sim/IsaacLab
Robot learning framework built on Isaac Sim (Omniverse/PhysX); GPU-parallel simulation powers large-scale RL and Sim2Real training β the core of NVIDIA's embodied AI ecosystem.
huggingface/lerobot
Real-robot learning framework from Hugging Face: ACT/Diffusion Policy/VLA algorithms out of the box, paired with low-cost hardware like SO-100 and a dataset ecosystem; the most active community in the field.
Genesis-Embodied-AI/Genesis
General-purpose robot simulation platform driven by a generative physics engine, focused on extreme simulation speed (~100x faster training); covers differentiable physics, generative scenes and manipulation policy learning.
openvla/openvla
7B-parameter open-source VLA foundation model (Prismatic VLM trained on Open X-Embodiment data); fine-tunable on a single GPU and widely used as an embodied-model baseline.
NVIDIA/Isaac-GR00T
NVIDIA's open-source humanoid foundation model GR00T N1 (dual-system architecture: fast VLM system + slow diffusion action blocks), with data pipelines and fine-tuning scripts.
thu-ml/RoboticsDiffusionTransformer
Largest open-source bimanual manipulation diffusion foundation model (1B params, jointly trained across multiple robot embodiments); led by a Chinese team.
google-deepmind/open_x_embodiment
Million-scale cross-institution robot manipulation dataset (22 embodiments, 1M+ trajectories); the most important training-data cornerstone of the VLA era.
droid-dataset/droid
Large-scale diverse robot manipulation dataset (350 hours, 564 scenes, 86 tasks), designed for policy generalization research.
PKU-EPIC/DexGraspNet
Large-scale dexterous grasping pose dataset (1M+ grasps, URDFs for multiple dexterous hands); a common benchmark for dexterous manipulation research.
tonyzhaozh/aloha
Open-source implementation of the low-cost bimanual teleoperation hardware ALOHA and the imitation-learning policy ACT; sparked the low-cost bimanual robotics research ecosystem (precursor of Mobile ALOHA).
leggedrobotics/legged_gym
RL training framework for legged robots built on Isaac Gym (paired with rsl_rl); the classic starting point for legged and whole-body-control Sim2Real research.
ros2/ros2
Robot Operating System 2 (middleware + toolchain + ecosystem standards); the de facto software standard for real robots and the runtime base for most embodied AI deployments.
OpenLoong/OpenLoong
Open-source full-stack humanoid robot platform (Qinglong robot), covering motion control, simulation training and full-system implementation; a flagship of China's humanoid open-source ecosystem.
haosulab/ManiSkill
General-purpose robot manipulation simulation and benchmark platform on SAPIEN (ManiSkill2/3), with GPU-parallel rendering and evaluation; rich in manipulation tasks.
haosulab/SAPIEN
Physics-based articulated-body simulation platform strong in articulated-object interaction and differentiable rendering; the underlying base of ManiSkill.
ARISE-Initiative/robosuite
Modular robot-arm manipulation simulation framework on MuJoCo (paired with robomimic), with standardized tabletop tasks and imitation-learning interfaces.
StanfordVL/OmniGibson
Large-scale interactive household-scene simulation on Isaac Sim (base of the BEHAVIOR benchmark); emphasizes realistic interaction semantics and long-horizon tasks.
Lifelong-Robot-Learning/LIBERO
Lifelong robot learning benchmark (130 tasks, 5 generalization axes); one of the most cited manipulation benchmarks in VLA papers.
mees/calvin
Language-conditioned long-horizon manipulation benchmark (PYMO-Sense base + simulated sensors); representative for multi-round sequential task evaluation.
stepjam/RLBench
100-task vision manipulation benchmark on CoppeliaSim; a classic manipulation RL/IL evaluation suite.
isaac-sim/IsaacGymEnvs
GPU-parallel RL environment set shipped with Isaac Gym Preview (Ant/ShadowHand/Factory etc.); merged into Isaac Lab but still widely used in papers.
gazebosim/gz-sim
Long-established open-source 3D robot simulator (new-generation gz-sim), deeply integrated with the ROS ecosystem; a standard for industrial and research robot simulation.