Start Embodied AI Training Faster

Simaihub builds structured, training-ready simulation data for embodied AI — designed for reinforcement learning, offline RL, and imitation learning.

Built on NVIDIA Isaac Sim and Isaac Lab, our data products plug directly into real training workflows, so robotics teams can move from simulation to model iteration faster.

Our flagship product is a four-scene mobile robot navigation training dataset for AMR / mobile robots, generated on NVIDIA Isaac Sim and Isaac Lab.

• Scenes: Hospital, Office, Warehouse, Simple Room
• Format: observations, actions, rewards, dones, episode metadata (HDF5)
• Coverage: diverse routes, dynamic obstacles, recovery / unstuck behaviors, lighting & sensor randomization
• Use cases: RL / offline RL, imitation learning, sim-to-real research

We also support custom datasets by robot type, scene, sensors, and training objectives — plus optional Isaac Lab agent training services.

• Training-first data design for RL / offline RL / IL — not raw log dumps
• NVIDIA-native stack: Isaac Sim + Isaac Lab for scalable, physically based collection
• Generalization-oriented coverage: multi-scene, dynamics, recovery, domain randomization
• Standard products + custom enterprise programs

• Expanded navigation scenes and robot platforms
• Robotic arm / manipulation training datasets
• Multi-sensor embodied training data
• Custom simulation data programs for enterprise and research teams

Need simulation data for embodied AI training, or a custom dataset for your robotics workflow? Get in touch.