Galbot Launches LDA-1B World-Action Model, Open-Sources Framework

Galbot has launched the LDA-1B model, a foundational model comprising 1.6 billion parameters. This model is based on the company's proprietary WAM architecture, integrating world models and work models at the data level, enabling shared learning from both real and simulated data. The training process utilized between 5,000 to 30,000 hours of data, significantly reducing prediction error in operations. The research on this model has been accepted for presentation at the RSS 2026 conference, and the source code has been made publicly available.