# Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation

**Authors:** Zipeng Fu, Tony Z. Zhao, Chelsea Finn  
**arXiv:** [2401.02117](https://arxiv.org/abs/2401.02117)  
**Date:** 2024-01-04  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2401.02117v1/

## Key points

- Imitation learning from human demonstrations has shown remarkable performance in table-top manipulation tasks
- Results often lack the mobility and dexterity required for practical and versatile applications
- Mobile ALOHA is a system developed by Zipeng Fu, Tony Z. Zhao, and Chelsea Finn to enable imitation of bimanual mobile manipulation tasks with whole-body control
- The system incorporates a low-cost whole-body teleoperation interface and adds a mobile base to enhance capabilities
- Co-training with existing static ALOHA datasets significantly improves performance on mobile manipulation tasks
- Up to 90% success rates achieved with 50 demonstrations for each task
- Mobile ALOHA can autonomously complete complex mobile manipulation tasks such as sautéing, serving, opening cabinets, calling an elevator, and rinsing pans using a kitchen faucet
- Presents an innovative solution for imitating bimanual mobile manipulation tasks that require whole-body control
- Opens up possibilities for more advanced robotic applications in various domains

## Summary

In the field of robotics, imitation learning from human demonstrations has shown remarkable performance in table-top manipulation tasks. However, these results often lack the mobility and dexterity required for practical and versatile applications. To address this limitation, a team of researchers consisting of Zipeng Fu, Tony Z. Zhao, and Chelsea Finn have developed Mobile ALOHA - a system that enables the imitation of bimanual mobile manipulation tasks with whole-body control. The system incorporates a low-cost whole-body teleoperation interface and adds a mobile base to enhance the existing ALOHA system's capabilities. By utilizing this system to collect data, the researchers then perform supervised behavior cloning. Their findings indicate that co-training with existing static ALOHA datasets significantly improves performance on mobile manipulation tasks, with up to 90% success rates achieved with 50 demonstrations for each task. This enhancement allows Mobile ALOHA to autonomously complete complex mobile manipulation tasks such as sautéing and serving a piece of shrimp, opening a two-door wall cabinet to store heavy cooking pots, calling and entering an elevator, and lightly rinsing a used pan using a kitchen faucet. Overall, the development of Mobile ALOHA presents an innovative solution for imitating bimanual mobile manipulation tasks that require whole-body control and opens up possibilities for more advanced robotic applications in various domains.