In this report, we present Hy-Embodied-0.5-VLA, abbreviated as HyVLA-0.5, an end-to-end system that spans the full robot learning stack: data collection, model design, continued pre-training and supervised fine-tuning, RL post-training, and real-world deployment. Each component serves a distinct role in this stack.

翻译：在本报告中，我们介绍 Hy-Embodied-0.5-VLA（简称 HyVLA-0.5），这是一个覆盖完整机器人学习流程的端到端系统：数据采集、模型设计、持续预训练与有监督微调、强化学习后训练以及真实世界部署。该系统中的每个组件均承担着独特的作用。