Wall-OSS-0.5, an open-source embodied intelligent model for independent variable robots.

X Square Robot's open-source embodied large model, Wall-OSS-0.5, allows for pre-training and deployment. In zero-shot tests on 17 real-world robotic tasks, it scored over 80 points in four categories. The model is pre-trained on over 20 robot forms, over one million trajectories, and 90 million multimodal data points. It employs gradient bridging collaborative training and a visually aligned action tokenizer. After fine-tuning, its task performance surpasses the industry benchmark π0.5 by 17.5 points.