Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it.

AI robotics companies are outsourcing the collection of physical training data to specialized firms like XDOF, acknowledging the unglamorous and labor-intensive nature of this task. This data is crucial for advancing physical AI capabilities to match the progress seen in large language models (LLMs). XDOF, for instance, is already engaged in collecting this data for several AI labs, indicating a growing trend in the industry. The process involves meticulously gathering information from real-world interactions, which is essential for robots to learn and perform complex physical tasks. Without this extensive and detailed physical data, the development of sophisticated robots capable of nuanced manipulation and navigation will be significantly hindered. The reliance on external companies like XDOF highlights the challenges and costs associated with creating the datasets necessary for training advanced physical AI systems, a bottleneck that needs to be addressed for the field to progress.