According to a recent LinkedIn post from X Square Robot, the company has collaborated with researchers from Tsinghua University and Beijing Institute of Technology to open-source a system called HOST (Human-to-robot One-Shot Skill Acquisition). The post describes HOST as enabling robots to learn new manipulation skills from a single human demonstration video, averaging 29 seconds, without task-specific fine-tuning or parameter updates.
The company’s LinkedIn post highlights that HOST does not replicate human motion directly but instead aligns robot execution with task progress, predicts the demonstrated outcome from the robot’s perspective, and derives the necessary actions to achieve that outcome. The system was reportedly evaluated on 50 novel manipulation tasks and, on an eight-task benchmark, achieved a 62% average success rate, which the post indicates is 43 percentage points higher than the strongest compared method without parameter updates.
For investors, the post suggests that X Square Robot is positioning itself at the forefront of data-efficient robot learning and manipulation, an area of interest for industrial automation, logistics, and advanced manufacturing applications. Open-sourcing HOST via project materials, GitHub, and Hugging Face may help accelerate ecosystem adoption, attract academic and developer engagement, and potentially strengthen the company’s technology brand and partnership pipeline, although direct monetization pathways are not detailed in the post.


