Unitree Releases UnifoLM-OminiA-0.3 for G1 Humanoid Robot
Unitree has released the UnifoLM-OminiA-0.3 model, enabling its G1 humanoid robot to perform fully autonomous household and caregiving tasks. The model supports multi-modal interaction and closed-loop control, allowing the robot to handle object manipulation and device control with anti-interference capabilities. This release marks a significant step forward in embodied AI and general-purpose robotics by demonstrating a single model coordinating complex, multi-modal household tasks. It showcases the practical application of Vision-Language-Action (VLA) concepts in consumer-accessible humanoid hardware like the G1. In demonstration videos, the G1 robot successfully performed tasks such as sorting medicine boxes, loading a dishwasher, and adjusting a smart bed's height, which immediately stopped upon hearing the voice command "stop." The model integrates perception and action to achieve closed-loop control, though detailed technical specifications remain limited in the announcement.
## BACKGROUND
Unitree Robotics is known for developing affordable humanoid robots, such as the G1, which features advanced motor configurations and force-position hybrid control. The company's UnifoLM series, including previous models like UnifoLM-VLA-0, focuses on Vision-Language-Action architectures designed to bridge the gap between AI perception and physical robot manipulation.