MM-ABC: Towards Generalist Mobile Manipulation via Seeing, Coordinating and Imagining
MM-ABC is a robot foundation model built to coordinate mobile bases and arms as one system.
The paper says the model combines sparse multi-level VLM features, a training-only future-imagination branch, and masked joint attention for arm-base coordination. It was pretrained on 5,000+ hours of robot data across 400,000+ episodes, 12 datasets, and 17 embodiments. In reported tests, it reaches 61.2% success on RoboCasa365, 99.1% on LIBERO, and 83% mean success across five real-world mobile manipulation tasks. HF Daily Papers' note
The paper says the model combines sparse multi-level VLM features, a training-only future-imagination branch, and masked joint attention for arm-base coordination. It was pretrained on 5,000+ hours of robot data across 400,000+ episodes, 12 datasets, and 17 embodiments. In reported tests, it reaches 61.2% success on RoboCasa365, 99.1% on LIBERO, and 83% mean success across five real-world mobile manipulation tasks. HF Daily Papers' note
score 4