Event

HKU and Yisheng Technology Release DexHoldem Real-Robot Benchmark

1 reports · 1 sources

The University of Hong Kong and Yisheng Technology team released the DexHoldem benchmark, bringing Texas Hold'em onto real robots, using a UR10e robotic arm and a Shadow dexterous hand at the same real card table to simultaneously evaluate dexterous manipulation, table-state perception, and closed-loop operation. The research has been accepted to the NeurIPS 2026 Workshop. The experimental platform consists of a UR10e robotic arm and a Shadow dexterous hand, with 105 demonstrations collected per task, of which 100 are used for training and 5 for validation. The system experiments fixedly use GPT 5.5 as the perceiver and main agent, and π₀ as the low-level policy; some tasks provide annotated historical states as context, while another 12 tasks provide no history.