Knowin unveils GLOW framework for one-shot robot learning

Knowin published a technical report on GLOW, a framework that lets household robots learn a new task from a single human demonstration and reuse the skill across objects, environments and tasks without retraining. On 18 simulated RoboDojo tasks GLOW averaged 62.2% success versus 22.2% for a GPT-6-based baseline.

• GLOW learns a skill from one demo video with model weights frozen
• RoboDojo: 62.2% success vs 22.2% for the GPT-6 baseline
• 86.7% average across six LIBERO-Pro perturbation conditions
• KnowinBrain-1.5 tops Embodied Arena's 2D QA leaderboard of 20 models