Synthetic Manipulation Dataset
Procedurally generated manipulation trajectories from physics simulators with perfect state information for scalable robot policy pre-training.
Dataset at a Glance
Comparison with Public Datasets
How Claru's dataset compares to publicly available alternatives.
| Dataset | Clips | Hours | Modalities | Environments | Annotations |
|---|---|---|---|---|---|
| RLBench | 100K | ~50 | RGB-D | CoppelaSim | Actions, keypoints |
| ManiSkill2 | 200K | ~100 | RGB-D | SAPIEN | Actions, rewards |
| Claru Synthetic | 200K+ | 1,000+ | RGB, Depth, PC | MuJoCo, Isaac | Perfect state, forces, rewards, contacts |
Use Cases
Policy Pre-Training
Massive pre-training in simulation before real-world fine-tuning. Example models: Octo, RT-X, OpenVLA.
Reward Model Training
Training reward models on synthetic success/failure examples. Example models: VIP, R3M, LIV.
Curriculum Learning
Progressively harder task configurations for staged policy training. Example models: ManiSkill, Isaac Gym, RoboCasa.
Key References
- [1]James et al.. “RLBench: The Robot Learning Benchmark.” IEEE RA-L 2020, 2020. Link
- [2]Gu et al.. “ManiSkill2: A Unified Benchmark for Generalizable Manipulation Skills.” ICLR 2023, 2023. Link
- [3]Makoviychuk et al.. “Isaac Gym: High Performance GPU-Based Physics Simulation.” NeurIPS 2021, 2021. Link
How Claru Delivers This Data
Claru generates synthetic manipulation data using MuJoCo and Isaac Sim with procedural scene randomization. Combined with Claru's real-world data, this enables hybrid training pipelines that maximize both scale and realism.
Frequently Asked Questions
MuJoCo and NVIDIA Isaac Sim with procedural scene generation for domain randomization.
Textures, lighting, object poses, camera angles, and physics parameters are randomized per episode.
Yes. Our format pipeline ensures synthetic and real data share identical action spaces and observation formats for seamless co-training.
Request a Sample Pack
Get a curated sample of synthetic manipulation data with full annotations to evaluate for your project.