AI

MM-ABC: Towards Generalist Mobile Manipulation via Seeing, Coordinating and Imagining

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 16:18:24 UTC.

Briefings

MM-ABC is a foundation model for mobile manipulation that integrates sparse multi-level VLM features with a training-only future branch for world imagination and geometric intent supervision. It coordinates separate manipulation and mobility streams via masked joint attention and clean-action x-prediction. Pretrained on 5,000+ hours of heterogeneous robot data across 12 datasets, the model achieves 44.71% success on EBench, 61.2% on RoboCasa365, and 83% mean success on five real-world tasks.

Key points

• MM-ABC combines sparse multi-level VLM features for spatial perception with a training-only future branch using world imagination and geometric intent as extra supervision.
• The model coordinates separate manipulation and mobility streams through masked joint attention and clean-action x-prediction, achieving 83% mean success on five real-world tasks.

Context

Mobile manipulation extends robot interaction beyond a fixed kinematic workspace by making the reachable region itself controllable. Existing approaches often decouple mobility and manipulation into separate action streams, but MM-ABC argues for efficient cross-stream collaboration through unified representations.

Limitations

Results are from controlled ablations and pretraining on 5,000+ hours of heterogeneous data; performance on real-world tasks without perturbation training is reported as 83% mean success across five tasks.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 22/10/2026, 06:56:08 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeicv6aykgqkkexhqszlfhqmqq75v6x3fhvieexoz5bbblfsrp3pmoq