2026 Tencent Kaiwu AI Global Open Held in Chengdu, Focuses on Quadruped Robots and Drones

On August 13, 2026, the "2026 Tencent Kaiwu AI Global Open and the 17th China University Student Service Outsourcing Innovation and Entrepreneurship Competition AI Special Category (Category D) National Final" was held in Chengdu. The event attracted 4,433 teams and 13,428 students from 666 universities. After months of competition, 21 teams entered the finals, competing on two physical tracks: quadruped robots and drones. Robotic dogs navigated complex terrain including slopes, steps, and gravel using their own visual perception, while drones were required to autonomously complete takeoff, obstacle avoidance, and hovering. Participants faced a core challenge in global embodied AI: migration from simulation to reality.

The quadruped robot reinforcement learning challenge was co-launched by Tencent Kaiwu and Unitree Technology. The final course was a 15-meter-long, 4.8-meter-wide one-way straight track featuring public obstacles such as slopes, tray steps, and gravel surfaces, plus a hidden challenge unit revealed on the day. Participants' models had to transfer directly from simulation to the real Unitree Go2 robot to complete all tasks, without a remote control, with camera feed input directly to the model, which performed visual understanding and real-time motion control of 12 joints using the robot's own computing power, without relying on cloud or external networks. After the event, a student representative from Shenzhen University noted that migration from simulation to real robots is very difficult, as simulation did not fully account for mechanical structure, ground noise, or other interference. To enhance the model's functional stability in the real world, the team proactively added various interference and noise to the training environment.

The drone competition, part of the low-altitude intelligence track, was co-launched by Tencent Kaiwu, Tsinghua University, and Chengdu Thunderbird Technology. Drones were required to take off fully autonomously, plan their own routes, navigate obstacles, and ultimately hover. Competitors found that obstacles such as pillars in the real environment made localization extremely difficult, a scenario not encountered in simulated training. The head of industry-education integration at Unitree Technology stated that changes in physical friction, sensor noise, hardware latency, and terrain uncertainty are difficult to replicate fully in simulation, and that every step of exploration by participants is paving the way for the industry, with real "failure data" more valuable than perfect simulation results, helping the industry clarify the boundary between simulation and reality and identify breakthrough directions.

During the event, Unitree Technology's technical experts and executives observed team performance and technical solutions, and extended job offers to some outstanding participants. Tencent Kaiwu noted that companies can directly observe participants solving real problems, and since 2025, on-site recruitment has been held. Data indicate a current talent gap of approximately 1 million in embodied AI, and from January to April 2026, the embodied AI hiring index surged 15 times compared to the same period in 2025. The average annual salary in the humanoid robot track was RMB 406,100, but supply lagged behind, with only a limited number of universities offering undergraduate programs in embodied AI nationwide in 2026. The general manager of Tencent's Chengdu office said that AI is transitioning from tool to partner, and future AI talent must understand algorithms, hardware, and engineering. The competition aims to translate real industry needs into contest tasks, helping students complete a full cycle from theory to practice.

The event created a chain linking classroom learning, platform training, competition validation, research exploration, and industrial practice. The Tencent Kaiwu platform, incubated in Chengdu, benefits from the city's abundant university resources and diverse application scenarios, such as forest and agricultural environments. The platform aims to connect academia with industry, further integrating talent cultivation with real-world scenarios to develop skilled professionals aligned with industrial growth.