从进厂到马拉松：人形机器人离“实用”还有多远？

Core Insights - The first global "human-robot marathon" concluded, highlighting the capabilities of humanoid robots in a challenging environment, with only 6 out of 20 teams completing the race [1][5] Group 1: Performance and Competition - The marathon covered a distance of 21.0975 km, featuring various complex terrains that tested the robots' joint coordination, control algorithms, energy management, and structural design [1] - The winning team, TianGong, completed the marathon in 2 hours, 40 minutes, and 42 seconds, while the second and third places were taken by teams using the N2 and the "XingZhe ErHao" robots, respectively [1][2] - Performance differences were noted among the top three teams, with TianGong's robot achieving a high pace due to its height and weight advantages, while the N2 focused on stability and the XingZhe ErHao utilized a walking strategy [2] Group 2: Technological Insights - TianGong's robot employed a lightweight design using carbon fiber and a unique leg structure to enhance stability and reduce joint impact during running, which contributed to its performance [3] - The N2 robot from Songyan Power utilized deep reinforcement learning and a hierarchical model for motion planning, allowing it to adapt in complex terrains [3] - The G1 robot from Yuzhu Technology faced criticism for its stability during the race, raising questions about its practical applications despite previous promotional successes [4] Group 3: Industry Challenges and Future Outlook - The marathon exposed several issues within the humanoid robotics industry, including overheating joints, insufficient balance, and the need for frequent battery changes [5] - Analysts suggest that the industry must address these technical shortcomings to enhance the viability of humanoid robots in practical applications [5][6] - The year 2025 is anticipated to be pivotal for humanoid robot mass production, potentially leading to significant advancements in data collection and training, which are crucial for overcoming current limitations [6]