AgiBot spotlights Expedition A3, showcasing production model’s achievements in martial arts, table tennis and ballroom dancing

AgiBot Innovation (Shanghai) Technology Co., Ltd. on Oct. 16, 2026, published “Behind the Gold Medals | Expedition A3: Master of Masters, All-Around Champion” on its website. The article summarizes Expedition A3’s competition and deployment examples, against the backdrop of AgiBot winning 18 gold, 16 silver and 12 bronze medals at the Second World Humanoid Robot Games and topping both the gold-medal and overall medal tables. The company said the Expedition A3 is a production model that demonstrated movement and coordinated control in different events, including tai chi, table tennis, kickboxing and ballroom dancing. It also said Expedition A3 robots are in regular service alongside hundreds of “silicon-based companions” at a large embodied AI theme park that opened Sept. 24 at Hengqin Chimelong Spaceship Park.

According to AgiBot, the Expedition A3 is a production model, and the company plans to incorporate algorithms and high-difficulty movements developed in competition into subsequent production versions. The competition results therefore matter as a link between one-off performances, real-world operations and product updates. The article also describes regular operations at Hengqin Chimelong Spaceship Park, suggesting robots are being assessed not only for their performance in demonstrations but also for sustained use in cultural and entertainment performances, educational programs and hotel services.

The key point is that AgiBot is repositioning the Expedition A3 not as a “competition robot” but as an example of performance validation and expanded applications for a production model. The article emphasizes medal counts and the number of events, but its focus appears to be less on the results themselves than on a single production model handling different control demands: high-difficulty spins in tai chi, the perception-prediction-control sequence in table tennis, and multi-robot coordination in ballroom dancing. In other words, the example is about more than awards; it connects hardware capabilities, motion control, content creation and real-world operations. In a related article, AgiBot delivers its 20,000th unit to Chimelong Group; more than 300 units operating at the park, this newspaper reported that more than 300 units were deployed across seven settings at Hengqin Chimelong Spaceship Park. AgiBot’s latest article takes that deployment as its starting point and frames the Expedition A3’s evolution from a “silicon-based star on stage” to a “24-hour productive partner.” While the earlier article highlighted the scale of deployment, this one uses competition results to lend technical justification to the robots’ expansion into those settings. At the industry level, the issue is not just a competition over individual robot performance. Tai chi tests high-dynamic motion control; table tennis requires shorter delays from perception to decision-making; kickboxing raises questions of impact resistance and safety; and ballroom dancing requires synchronization among multiple robots. The focus is whether production models can meet these varied demands. The article also places theme-park operations at Hengqin alongside the participation of academic and educational teams, presenting real-world use, education partnerships and motion generation as parts of a single cycle. Open questions include the number of Expedition A3 units produced, success rates and uptime in each event, the robots’ specific roles at the theme park, and how the “generative transition strategy” and multi-robot coordination mentioned in the article will be reflected in production versions. How much of the competition-developed motion will make it into product specifications, and how consistently it can be reproduced in cultural tourism and guest-service settings, remain to be seen.

For tai chi, the article describes a sequence of “outside crescent kick 450°,” “outside crescent kick 360° into horse stance” and “outside crescent kick 540° into horse stance,” saying the Expedition A3 performed high-difficulty spins. In kickboxing, it said the “Shangxi GOGO Team,” which included students from the Shanghai Theatre Academy, Xi’an Jiaotong-Liverpool University and the China Academy of Art, took part and collected more than 10 hours of martial arts movements using optical motion capture for simulation and reinforcement learning. For table tennis, the article said a joint team from Peking University and the Beijing Academy of Artificial Intelligence won gold, and that the Expedition A3 demonstrated an integrated process involving visual recognition, trajectory prediction, strategy decisions and movement execution. In ballroom dancing, it said two Expedition A3 robots had to coordinate simultaneously, and that solving issues involving multi-robot communication and motion synchronization led to a gold medal.

In China, production-model performance is being tested across competitions, theme-park operations and education partnerships. For Japanese universities, research institutions and cultural tourism facilities pursuing similar demonstrations, the question is likely to be whether perception, control and content creation can be connected to sustained operations rather than one-off demonstrations.