GEN-0

GEN-0 is Generalist's predecessor embodied foundation-model generation, trained on large-scale real-world manipulation data and designed to transfer across robot embodiments.

Facts

  • Company: Generalist
  • Type: Robotics Foundation Model
  • Positioning: Physical-interaction pretraining
  • Function: Learn manipulation policies across embodiments
  • Audience: Embodied-AI and robotics teams
  • Status: Historical
  • Website: https://generalistai.com/blog/gen-0

About

About GEN-0

GEN-0 is Generalist’s embodied foundation-model generation for multimodal training on raw physical interaction. Harmonic Reasoning interleaves asynchronous, continuous-time streams of sensing and acting tokens, allowing the model to coordinate perception and control across different robot embodiments.

Generalist tested GEN-0 on 6-DoF, 7-DoF and 16-plus-DoF semi-humanoid robots and scaled the family beyond 10 billion parameters. The company reports pretraining on more than 270,000 hours of real-world manipulation data spanning homes, warehouses and workplaces, with downstream tasks including assembly, packing, clothing handling and food preparation.

Specifications

  • Reported Model Scale: Scaled beyond 10 billion parameters
  • Tested Embodiments: 6-DoF; 7-DoF; 16+ DoF semi-humanoid robots
  • Pretraining Data: More than 270,000 hours of real-world manipulation data
  • Data Growth Rate: More than 10,000 new hours per week
  • Reasoning Architecture: Harmonic Reasoning with asynchronous continuous-time sensing and acting tokens