Embodied AI Moves From Competition Arena to the Shop Floor

Highlights

• Ant Lingbo embodied challenge names top 15; on-hardware final set for Nov 14 in Shanghai
• DEEP Robotics X50 launched: 50 kg payload, 3 m/s, 2 km range
• Apple's Turkey advanced manufacturing center to open in 2028, with robotics training

Embodied Large Models: The Competition Enters On-Hardware Validation

The finalist list for the first Ant Lingbo Embodied Large Model Challenge was released on Nov 6. Fifteen teams and 59 contestants advanced from 1,855 teams and 3,252 contestants worldwide. The event covers 13 countries and 648 universities and research institutions. GPU allocation for the preliminary round was raised from a planned 200 to 350, and the two stages together provided more than 340,000 GPU-hours of compute (Source: Leiphone).

The final format matters more: on Nov 14–15, the 15 teams will hold a 48-hour offline on-hardware hackathon in Zhangjiang, Shanghai. All will use the Robbyant Forge developer platform and AgileX COBOT MAGIC devices, building on the open-source LingBot-VLA 2.0 pretrained model to complete the full workflow from task understanding and data collection to model tuning and on-hardware deployment (Source: Leiphone). That means the judging criteria shift from task success rates in the preliminary round to model generalization and hardware-software engineering capability.

Beyond the event, Ant Lingbo disclosed that its open-source models have accumulated more than 33,000 GitHub stars, and that LingBot-VLA and LingBot-VA have been adapted to LeRobot. A developer conference on Nov 16 will announce a developer ecosystem plan (Source: Leiphone). The combination of open-source models and on-hardware competition is pulling embodied AI competition back to reproducible engineering validation.

Quadrupeds: Heavy Payload and Mobility Are No Longer Either-Or

DEEP Robotics launched the X50 quadruped on Nov 6, raising continuous-travel payload to 50 kg without increasing body size. Official figures show that under a rated 50 kg payload, the robot can run continuously at 3 m/s for more than 2 km, and can repeatedly climb 80 cm platforms, cross 70 cm gaps between stone blocks, and climb open-riser stairs (Source: Leiphone).

On the technical side, the X50 uses a perception-control fusion approach: it fuses lidar point clouds and IMU data in real time, uses reinforcement-learning motion control to predict terrain and plan footholds, and executes with high-stiffness, high-response joints (Source: Leiphone). DEEP Robotics explicitly targets power inspection, firefighting and emergency rescue — for example, carrying hydraulic tools or insulator strings, or hauling fire hoses for sustained spraying.

Commercialization of quadrupeds has long been constrained by the physical trade-off: add payload, lose mobility. The X50 puts heavy payload and all-terrain capability on one machine. If it can reliably reproduce that performance in real settings such as power-plant boiler rooms and chemical pipe galleries, its value will go beyond the spec sheet.

Manufacturing and Talent: Robotics Enters Industrial Training

Apple plans to set up seven training facilities in Turkey, including an Ankara advanced manufacturing center due to open in 2028. It is Apple's second advanced manufacturing center and the first of its kind outside the US. Once built, it will offer free consulting and training to small and medium-sized enterprises, covering AI, machine learning, predictive maintenance, automation and robotics (Source: ITHome).

This thread does not involve robot products directly, but it sends a signal: robotics is being folded into the infrastructure of manufacturing workforce training. Apple's first advanced manufacturing center opened in Houston, Texas, in August this year. The Turkey center's curriculum shows that automation and robotics have become general needs for manufacturing upgrades, not a topic reserved for a handful of leading companies.

Other News

• Roborock's P30 standard water-tank version robot vacuum is listed, with 25,000 Pa suction and a wash-free, self-emptying design, priced at 2,999 yuan (Source: ITHome).
• Graphite research shows ChatGPT accounts for more than 81% of global AI chatbot traffic. The figure belongs to the AI application layer and is not directly related to robot hardware (Source: ITHome).