π0 and π0-FAST: Vision-Language-Action Models for General Robot Control

The article introduces two new models, π0 and π0-FAST, designed for vision-language-action tasks in general robot control. π0 utilizes a transformer-based architecture with a multimodal input that integrates visual and linguistic data, while π0-FAST optimizes for efficiency, achieving real-time performance with reduced computational overhead. These models enhance the ability to train robots in complex environments using natural language instructions, which is critical for advancing human-robot interaction and autonomous task execution in practical applications.