Zioneer Robot Team

“By applying their GRPO post-training framework, the authors improved the average success rate on 24 standard RoboCasa manipulation tasks from 67.4% to 80.6%. As stated in Section 4.2, 'Z-1 RL improves the average success rate from 67.4% to 80.6%, corresponding to a gain of 13.2 percentage points.'”

“Z-1 achieves state-of-the-art results on the RoboCasa benchmark using only publicly available data and a publicly available base model ($\pi_{0.5}$). The authors state in the Abstract: 'These results show that systematic GRPO post-training can substantially improve flow-based VLA policies without additional private demonstrations.'”