The Best New Tech at WAIC 2026 | Humanoid Robots Review

In this talk

1. 0:00

   Robots move toward production work

   AGIBOT presented 60 robots across the expo, including machines for guiding visitors, making coffee, and working on a factory line. The company said its androids had completed six days on a tablet manufacturing line. Its A3 Ultra has 51 active degrees of freedom and used camera, lidar, and NVIDIA Thor computing to play table tennis without operator control.

2. 1:48

   Different designs target different jobs

   AGIBOT’s G2 Max can lift 18 kilograms with one arm, while its X2 Edu is designed for training and experimentation. P&D Bionics showed its human-sized Atom robot climbing stairs and tackling obstacles, with teleoperation supporting hand tasks as the company develops autonomous capabilities. Fourier Intelligence demonstrated a GR-3 responding to broad requests and coordinating with other robots, alongside smaller GR-Nano companions intended for older and lonely people.

3. 6:02

   Service robots tackle everyday tasks

   No Matrix demonstrated robots handling pharmacy orders and hotel laundry, including folding textiles. Its RoboPocket system gathered more than 100,000 hours of recordings from people in 47 Chinese cities, capturing depth and hand-object interaction to train robots on varied tasks. Keenon showed a multi-robot laundry workflow that took about four minutes and said its service robots already operate in more than 5,000 hotels.

4. 8:31

   Autonomy depends on learning and coordination

   Yunex AI showed a robot making customized pizzas and other machines preparing food, cleaning, and handling laundry. Xera AI demonstrated folding clothes and loading a washing machine, while Tencent presented Xiaolu, a massage robot trained on professional practitioners’ movements and applied force. These examples put flexible materials and safe physical interaction at the center of service robotics.

5. 10:32

   Computing infrastructure supports robot fleets

   Huawei introduced Atlas 950 SuperPod, a computing system rated at up to eight exaflops, built by linking accelerators through UnifiedBus. MiniMax presented its open-source M3 model, with 427 billion parameters and a context window of up to one million tokens. The demonstrations connected robot capabilities to larger systems for training, planning, and coordinating machines.