Action Tokenization
تكميم الأفعال
عملية تحويل أبعاد الفعل الروبوتي المستمرة (موضع، دوران، قابض) إلى عدد محدد من المستويات المنفصلة (عادةً 256)، ثم تمثيل كل مستوى كرمز نصي يستطيع نموذج اللغة توليده.
The process of discretizing continuous robot action dimensions (position, rotation, gripper) into a fixed number of bins (typically 256), then representing each bin as a text token that a language model can generate.
Also translated asترميز الأفعال، تقطيع الأفعال إلى رموز
First appears in this corpus in: A Generalist Agent (2022)
Appears in these papers
- A Generalist Agent2022in the sky ✦
- Open X-Embodiment: Robotic Learning Datasets and RT-X Models2024in the sky ✦
- Open X-Embodiment: Robotic Learning Datasets and RT-X Models2024in the sky ✦
- π₀: A Vision-Language-Action Flow Model for General Robot Control2024in the sky ✦
- RT-1: Robotics Transformer for Real-World Control at Scale2022in the sky ✦
- RT-1: Robotics Transformer for Real-World Control at Scale2022in the sky ✦
- RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control2023in the sky ✦
- RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control2023in the sky ✦