机械设备行业专题研究：机器人大规模训练开启，数据集和动作捕捉设备需求有望放量

Investment Rating - The report maintains an "Accumulate" rating for the industry [15]. Core Insights - The Shanghai government has released an implementation plan focusing on the importance of data collection in the "AI + Embodied Intelligence" sector, emphasizing the need for large-scale training data sets for humanoid robots by 2025 [20][21]. - The report identifies four primary methods for producing training data sets for humanoid robots: based on real bodies, virtual bodies, synthetic data, and human data, with a preference for real and virtual body methods due to quality concerns [20][24]. - The demand for motion capture devices is expected to increase significantly as more humanoid robot manufacturers enter the market, with each company potentially requiring hundreds of such devices [45][52]. Summary by Sections Industry Overview - The report highlights the critical role of data quality in humanoid robot training, indicating that poor data quality can hinder technological advancements [2]. - The implementation plan aims to establish a world-class AI industry ecosystem by the end of 2025, focusing on open-source technology and data collection [20]. Data Collection Methods - Real body data production involves using motion capture devices to collect visual, operational, and sensory information from humanoid robots, which is the highest quality but also the most expensive method [14][24]. - Virtual body data production, promoted by Nvidia, captures data in a simulated environment, providing high-quality data but facing challenges in transitioning between virtual and real scenarios [21][24]. Motion Capture Technology - The report discusses the workflow of motion capture systems, which includes data processing, output, body mapping, and platform adaptation, emphasizing the importance of body mapping algorithms [32]. - Major players in the motion capture market include Movella and Noitom, with significant technological barriers to entry [37][45]. Investment Recommendations - The report suggests focusing on companies with core technologies related to motion capture devices, recommending specific companies such as Aofei Entertainment for motion capture and Tianyu Shuke and Nengke Technology for data sets [52].