AIBRIEF
Kinetix ran its perception and control stack on a 117 joint humanoid at the 2026 World Robot Conference in Beijing. The 30 year arc of ping pong robots and the absence of peer review put that record on a booth.
At the 2026 World Robot Conference in Beijing, Shenzhen-based Kinetix showed a humanoid robot playing a full table tennis rally against human opponents and called it the "world's first autonomous humanoid table tennis complete match." The company's SMASH 2.0 system ties perception, trajectory prediction, motion planning, and whole-body control into a millisecond loop on a body the vendor describes as the highest-DoF humanoid in the world (LeiPhone; WRC exhibitor page).
The numbers behind the demo are dense: a 37-DoF hand with 20 active, 1 passive, and 16 compliant joints, 30N fingertip force, a roughly 300g head-mounted capture rig, and a data-and-training stack Kinetix says runs 10x faster on data production, 5x faster on training, and pushes models to robots in milliseconds (LeiPhone; humanoid.press profile). An alternative-angle piece from NetEase frames Kinetix as a case study in moving embodied-AI experience from human demonstration to robot action (163).
Table tennis is a long-standing stress test for whole-body control. Omron's FORPHEUS has been hitting rallies since the 1990s; DeepMind and Google have published reinforcement-learning work on the same problem. Against that arc, "first" is a moving bar. The LeiPhone report, driven by Kinetix's booth messaging, names no independent timing, no controlled conditions, and no peer-reviewed paper.
Until third-party timing, a named external expert, or a public paper lands, the "first" is a vendor claim, the 117-DoF figure is a vendor figure, and the rally is a booth demo.