📄 中文摘要
ABot-M0 是一个框架，旨在解决机器人领域中构建通用嵌入式代理的挑战，尤其是在多种硬件环境下的应用。该框架通过系统化的数据整理管道，联合优化模型架构和训练策略，能够将异构原始数据转化为统一且高效的表示。基于六个公共数据集，清理、标准化并平衡样本，构建了 UniACT 数据集，这是一个大规模数据集，包含超过 600 万条轨迹和 9500 小时的数据，涵盖了多种机器人形态和任务场景。
ABot-M0：用于机器人操作的 VLA 基础模型与动作流形学习
📄 English Summary
ABot-M0: VLA Foundation Model for Robotic Manipulation with Action Manifold Learning
ABot-M0 is a framework designed to address the challenges of building general-purpose embodied agents across diverse hardware in the field of robotics. It establishes a systematic data curation pipeline while jointly optimizing model architecture and training strategies, enabling the end-to-end transformation of heterogeneous raw data into unified and efficient representations. From six public datasets, samples are cleaned, standardized, and balanced to construct the UniACT dataset, a large-scale dataset comprising over 6 million trajectories and 9,500 hours of data, covering various robot morphologies and task scenarios.