Axis Robotics has released the Axis Sim Dataset V1, positioning it as one of the largest open-source simulation datasets for Franka arm manipulation. The full dataset, along with its training code and benchmarks, has been made publicly available.
Built using a simulated Franka Research 3 arm, the V1 dataset comprises more than 50,000 human-teleoperated simulation trajectories spanning 207 manipulation tasks and over 60,000 scene variants. The release has drawn over 160,000 downloads on Hugging Face, making it the most downloaded open-source simulation Franka manipulation dataset on the platform.
Testing a New Thesis on Data Quality
While standard robotics methodologies often prioritize filtering down to near-optimal, expert trajectories while discarding noisy data, Axis Robotics pursues a different approach. The company's thesis suggests that data quality exists at the distribution level. When a diverse crowd produces suboptimal or noisy trajectories with uncorrelated errors, the noise averages out during training to support a functional policy.
The V1 dataset puts this concept to public testing with tasks ranging from pick-and-place, stacking, and pouring to articulated-object manipulation and tool use. Data was collected via the browser-based teleoperation platform Axis Hub by a distributed crowd in collaboration with researchers from UC Berkeley, Johns Hopkins, and the University of Michigan.
Performance and Benchmarks
According to benchmarking results on LIBERO-Plus, continual pretraining on V1 improved the success rate of π0.5 from 83.9% to 88.8%, outperforming a volume-matched RoboCasa365 baseline by 37.3%. Performance scaled consistently as pretraining data increased from 25% to 100%, with the most significant improvements observed under camera, sensor-noise, and layout perturbations.
Axis Robotics reports that development on V2 is already underway, targeting 1.2 million trajectories across 1,200 tasks. The company previously raised $12 million in seed funding led by Hack VC, with participation from Nomad Capital, Pi Network Ventures, 10K Ventures, and angel investors.
