Chinese robotics company AgiBot has launched its newest AI model, GO-1, also known as Genie Operator-1. This innovative artificial intelligence system is designed to improve task execution in humanoid robots , using a Vision-Language-Latent-Action (ViLLA) model that optimizes visual understanding and robotic actions using real-world data and internet videos.
An innovative framework for autonomous robots
The GO-1 model is based on the ViLLA framework, an advanced architecture that integrates a Vision-Language Model (VLM) and a Mixture of Experts (MoE). This system enables robots to understand complex scenes and execute precise actions by analyzing heterogeneous data .
While previous models of autonomous robots relied on direct links between action, vision and language, GO-1 predicts latent action tokens, creating a bridge between perception and execution.
Check out how well the GO-1 works. Source: AgiBot
Unique capabilities of the Genie Operator-1
One of GO-1’s most notable features is its ability to learn from human demonstrations via internet videos , allowing the robot to generalize to new scenes and tasks with minimal data. Furthermore, the model is able to quickly adapt to new robot shapes and environments, making it a highly flexible solution for diverse applications.
The most relevant features of the GO-1 system include:
- Learning from Human Videos: GO-1 can learn from web content and real-life human demonstrations, improving its understanding of human actions.
- Generalization with little data: GO-1’s strong generalization capabilities allow it to adapt to new tasks and scenes with minimal data, even in zero-shot scenarios.
- Adaptation between different robots: This generalist robot policy model can be easily transferred between different types of robots and adapted to new forms of execution.
- Continuous Evolution: GO-1 adapts and evolves based on data generated during real-world use, making it a perfect tool for dynamic robotic intelligence.
Significant improvement in robot tasks
GO-1’s capabilities were validated on five tasks of varying complexity, where it outperformed the most advanced models on the market. On tasks such as “pouring water” or “replenishing drinks,” the model improved success rates from 46% to 78% .
These results demonstrate the effectiveness of the system in performing complex tasks with high accuracy, thanks to the integration of the Latent Planner and the Action Expert into the ViLLA framework.
An autonomous and generalist future
With the launch of GO-1, AgiBot is taking a step towards a future where autonomous robots will not only be tools for specific tasks, but autonomous agents with general intelligence. This model has the potential to transform multiple sectors, including manufacturing, services, and domestic applications .
Furthermore, GO-1’s continued evolution in the real world will enable robots to constantly adapt to new instructions and dynamic environments. Thus, AgiBot GO-1 accelerates the widespread adoption of embedded intelligence , paving the way for more versatile and advanced robotics in the near future.
Follow us on social media and don’t miss any of our posts!
YouTube LinkedIn Facebook Instagram X (Twitter) TikTok
Source and photo: AgiBot
