Humanoid robot scores first autonomous full table-tennis match at Beijing conference
The 2026 World Robot Conference opened in Beijing on Monday with a milestone that puts embodied AI on the radar of anyone who follows the field: a humanoid robot completed the world's first fully autonomous table-tennis match — not a scripted rally, but a real game with serving, receiving, and adaptive shot selection against a human opponent.
超维动力 KAI (Chaoweidongli), a full-stack embodied AI company, demonstrated its SMASH 2.0 system on a standard ping-pong table at the Beijing Yichuang International Convention and Exhibition Center. The system chains together high-speed visual perception, real-time trajectory prediction, motion planning, and whole-body control into a closed loop that operates within millisecond windows — the kind of latency budget that makes table tennis one of the hardest sports to automate. Unlike prior demos that showcased single skills in isolation, SMASH 2.0 handles the full game flow: autonomous serves, returns, and adaptive shot selection including long balls and short drops, all coordinated across the robot's joints without human teleoperation.
What makes this more than a party trick is the underlying architecture. KAI's world model — trained on millions of video examples in simulation before being transferred to the physical robot — learns to predict how objects behave in the real world, then uses that understanding to plan movements. The same algorithm runs across different robot bodies (the company calls this "one brain, many forms"), which suggests the system isn't hard-coded for one specific machine but is learning generalized motor skills. The humanoid platform itself, KAIBot, has 117 degrees of freedom across its entire body — the highest count publicly claimed for a humanoid — giving it the joint range to replicate human-like shoulder, wrist, and torso coordination during fast exchanges.
The demonstration is significant because table tennis sits at an extreme point on the dexterity-speed spectrum: the ball travels at several meters per second, spins change its trajectory mid-flight, and the robot has less than 200 milliseconds to decide how to respond. Solving that requires tight integration of perception, prediction, and control — exactly the stack that embodied AI companies need to crack before robots can handle unstructured real-world tasks like warehouse picking or home assistance. If KAI's approach scales beyond sport to industrial settings, it validates the "world model plus whole-body control" paradigm that several Chinese robotics firms are betting on.
The conference runs through August 23, with KAI also showcasing its KAI Hand — a 37-degree-of-freedom dexterous gripper — and KAI Halo, a first-person data-collection headset for training robots from human demonstrations.
What's the hardest physical task you'd trust a humanoid robot to attempt next? Tell us in the comments.
Sources: QbitAI · 2026 World Robot Conference
