Hitch Open's robots rally 142 times at table tennis, up from 17, on a test it wrote and opened

Two robots kept a table-tennis rally going for 142 shots on Nov. 7 at the China International Import Expo in Shanghai, with no remote control: they perceived the ball, decided and swung on their own. The previous record for the event was 17 rallies, the Shanghai Observer reported. A team led by Professor Gao Yue of Shanghai Jiao Tong University and the Shanghai Innovation Institute trained the model; Hitch Open, a Silicon Valley platform with academic roots at UC Berkeley, supplied the robot body, simulator, motion capture and scoring rules.

The record was the stage for the launch that matters. The same day Hitch Open open-sourced its first set of tools: the HOPE table-tennis foundation model, a benchmark and leaderboard, a simulation environment and a motion-capture dataset. Every team on the leaderboard uses the same body and submits openly, so only the algorithms differ. "Whoever defines the test defines the industry," said Tang Minqin, Hitch Open's global executive chairman. An eightfold jump on a record is what an early benchmark looks like, and whoever writes its rules gets to set the pace.

The posted code is written for one AgiBot humanoid

The code Hitch Open has posted on GitHub is the full HOPE stack for AgiBot's Expedition A3, a humanoid robot with 31 actuated joints. Its rules ban racket tracking: no reflective markers on the racket, the robot's hand or its wrist. Table tennis leaves a robot about 100 milliseconds to react, IT之家 noted, which is why the organizers pitch it as a limit test for embodied AI perception, decision and control.

Hitch Open has been building toward this. In 2025 the lab of co-founder Allen Y. Yang reached a 106-shot rally between a human and a robot, and HOPE was one of only two AI autonomous-decision events at the World Humanoid Robot Games in Beijing in August 2026, where ten universities competed.

The next mark it has set is one Hitch Open cannot certify alone. Its "2030 plan," launched at the Middle East stop of its circuit, aims to have autonomous robots from several brands take on the world's best human players by 2030. The code it has posted so far supports one brand.