Gasgoo Munich- According to Gasgoo Embodied Intelligence, GenRobot recently helped Simate advance its Physical RSI (Recursive Self Improvement) roadmap. Leveraging the AutoResearch system, Simate secured the top spot on the RoboDojo embodied intelligence leaderboard by employing a human-machine co-driving weak RSI paradigm.
For this project, GenRobot relied on its Data Foundation Model (DFM) — an end-to-end system for generating multimodal data and enabling continuous evolution. By recording real-world operations, reconstructing action trajectories, and aligning multimodal information, the company provided critical data infrastructure for Simate's model learning and iteration.
Focusing on the precision and completeness of real-world operational data, GenRobot achieved stable hand tracking accuracy of under 1 centimeter, meticulously reconstructing hand positions and movement trajectories. Additionally, using six-channel fisheye first-person video, the system rebuilt continuous full-body 3D human meshes with an average error of roughly 3 centimeters. This integrates hand manipulation with the coordinated movement of the torso and limbs into a comprehensive human expression.
On the hardware perception front, GenRobot has built a data acquisition product matrix covering the head, hands, and full body, centered around its proprietary cognitive core, Ego. The lineup includes bionic dual-finger and dexterous five-finger hands, six-channel fisheye lenses, controllers, and industrial-grade grippers, with cumulative orders across the entire product line surpassing 10,000 units. Its proprietary multi-camera perception matrix utilizes six 200-megapixel RGB cameras, achieving multi-device synchronization latency of under 1 millisecond.
Furthermore, GenRobot has expanded into tactile dimensions by incorporating magnetic tactile, piezoresistive, capacitive, fabric, and force feedback modalities. This allows for the recording of actions and force feedback during real-world operations.
Looking ahead, GenRobot plans to continuously iterate its full suite of data acquisition products. The company also aims to further refine the DFM architecture to better enhance performance across multimodal and multi-task data scenarios.
