MM-ABC Enables Mobile Manipulation by Solving Dynamic Perception and Coordination Challenges
让机器人学会“边走边干活”：MM-ABC如何解决移动操作的两大难题
Researchers from HKUST (Guangzhou) and Tsinghua University propose MM-ABC, a foundation model for general mobile manipulation. The system addresses dynamic spatial perception and arm-base coordination through clear action prediction, multi-layer visual feature injection, and a future expert module used during training. Trained on over 5000 hours of data including the self-collected MM-30 dataset, MM-ABC achieves a 44.71% success rate on EBench and 61.2% on RoboCasa365, outperforming baselines like π0.5 in both fixed-base and mobile manipulation tasks.
Source: Chinese robotics — Dataset and collection discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
32.8%
Reported trials
View original evidence
如果不采用清晰动作预测，而是坚持用速度预测，论文的消融实验显示，在RoboCasa365基准测试的复合任务上，成功率会从32.8%直接掉到29.2%，掉了3.6个百分点。
Open source S19Reported success rate
29.2%
Reported trials
View original evidence
如果不采用清晰动作预测，而是坚持用速度预测，论文的消融实验显示，在RoboCasa365基准测试的复合任务上，成功率会从32.8%直接掉到29.2%，掉了3.6个百分点。
Open source S19Reported success rate
24.9%
Reported trials
View original evidence
消融实验证明了这一点:去掉这个未来监督分支后，成功率从32.8%掉到24.9%，足足掉了7.9个百分点，这是论文所有消融实验里降幅最大的一项。
Open source S22
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
如果不采用清晰动作预测，而是坚持用速度预测，论文的消融实验显示，在RoboCasa365基准测试的复合任务上，成功率会从32.8%直接掉到29.2%，掉了3.6个百分点。
Open source S19
消融实验证明了这一点:去掉这个未来监督分支后，成功率从32.8%掉到24.9%，足足掉了7.9个百分点，这是论文所有消融实验里降幅最大的一项。
Open source S22
用五步采样（也就是机器人实际部署时用的采样步数）生成动作时，清晰动作预测的任务分配误差比速度预测低了14%。
Open source S17
5000多小时的"喂养"：数据从哪里来 一个模型再精巧，没有足够多样的数据喂养也白搭。
Open source S24
Source：Chinese robotics — Dataset and collection discovery · news.qq.com