X Square Robot Open-sources Embodied AI Model Wall-OSS-0.5
X Square Robot has open-sourced the embodied large model Wall-OSS-0.5, proposing that it is deployable after pre-training. In zero-shot testing across 17 real-world robot tasks, 4 tasks scored over 80 points. The model was pre-trained on over 20 types of robot morphologies, more than one million trajectories, and 90 million multimodal corpus entries. It employs gradient bridging collaborative training and a vision-aligned action Tokenizer. After fine-tuning, its task progress leads the industry benchmark π0.5 by 17.5 points.
X Square Robot has open-sourced the embodied large model Wall-OSS-0.5, proposing that it is deployable after pre-training. In zero-shot testing across 17 real-world robot tasks, 4 tasks scored over 80 points. The model was pre-trained on over 20 types of robot morphologies, more than one million trajectories, and 90 million multimodal corpus entries. It employs gradient bridging collaborative training and a vision-aligned action Tokenizer. After fine-tuning, its task progress leads the industry benchmark π0.5 by 17.5 points.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.