China: Human-machine 'co-creation' on show at World Robot Conference.

Shotlist Beijing, China - Aug 19, 2026 (CCTV Video News Agency - No access Chinese mainland) 1. Various of entrance to World Robot Conference 2026 exhibition hall 2. Poster of conference 3. Various of exhibition booth, visitors 4. Various of robot boxing presentation in progress at Unitree booth 5. Robotic arm grabbing bag of chips, scanning barcode 6. SOUNDBITE (Chinese) Zhang Zhanqi, algorithm engineer, DexForce (starting with shot 5): "Currently, humans need to teach the robot, but we cannot teach it every possible situation. We select typical cases that we believe are beneficial for the data, and through demonstration data from humans, the model explores patterns, for example, where the barcode is, or how to grip the upper part of a bottle to stand it upright. All of this is driven by data, and the model finds a 'general strategy' to execute." 7. Robots demonstrating work on phone packing line 8. Robot, person playing air hockey table game 9. SOUNDBITE (Chinese) Zhao Pan, marketing director, Astribot (starting with shot 8/ending with shot 10): "We aim to build robots that can adapt to complex, high-difficulty tasks in unstructured scenarios, including home environments, retail spaces, and sophisticated biopharmaceutical operations. Beyond that, we place greater emphasis on the future relationship between humans and robots as one of 'co-creation' rather than replacement. So right now, we are focusing on expanding applications in a broader range of everyday life scenarios." 10. Robot zipping up small backpack 11. Various of sign of robot shop, robot moving bottle from shelf to basket 12. SOUNDBITE (Chinese) Lin Fan, public relations manager, MagicLab (starting with shot 11/ending with shot 13): "We have always had a vision of co-creating across a wide range of scenarios. Because one robot on its own can hardly present its capabilities at work. So we need to work closely with partners and clients from all industries to help robots land in real-world applications and realize their operational capabilities." 13. Robot placing package on conveyor belt 14. Robotic hands on display 15. SOUNDBITE (English) Angel Zinsel, digital content creator from Spain: "It's amazing. I think the things that you can see here are really at the edge of robotics. So I think it's a good thing I came here." 16. Various of robots at exhibition 17. Exterior of exhibition hall, visitors Storyline Chinese robotics firms at the ongoing 2026 World Robot Conference in Beijing are shifting focus from building standalone machines to harnessing adaptive intelligence, with engineers demonstrating how robots can learn 'general strategies' from human guidance to broaden the potential scope of their use across more real-world scenarios and industries. The five-day conference opened in the Chinese capital on Wednesday, displaying the latest technological advancements in the global robotics industry, and bringing together over 300 exhibitors, a nearly 70-percent increase from last year, who are presenting more than 3,000 exhibits across four pavilions covering robotics applications in manufacturing, health care, retail, and emergency rescue, among other areas. While industrial robots and high-precision arms remain a staple of the event, this year's expo places a notable emphasis on AI-driven perception and adaptability, with robots that can see, grip and adjust to carefully handle everyday objects, from oddly shaped packages to soft-sided cartons. In one demonstration, a robotic arm equipped with an AI vision model proved its worth in a potential shopkeeper scenario. Developed by Shenzhen-based firm DexForce, it was seen picking up products which had been placed at a checkout counter and carefully repositioning them until the item's barcode was clearly readable to scan. The engineer behind the system explained that robots are not trained on every possible scenario, but rather on representative cases, meaning they are having to apply their own knowledge and adapt to new conditions. Through repeated human demonstration data, the model learns a 'general strategy' and applies it to new situations, enabling it to make the subtle adjustments necessary to pick up a bottle, a bag of chips or a medicine box. "Currently, humans need to teach the robot, but we cannot teach it every possible situation. We select typical cases that we believe are beneficial for the data, and through demonstration data from humans, the model explores patterns, for example, where the barcode is, or how to grip the upper part of a bottle to stand it upright. All of this is driven by data, and the model finds a 'general strategy' to execute," said Zhang Zhanqi, an algorithm engineer for DexForce. This concept extends far beyond the checkout counter, with a representative of another Shenzhen-start up named Astribot explaining how they are focusing on widening the potential scope