Mobile ALOHA: Stanford students develop a low-cost robot that cooks and cleans

Stanford Computer Science Professor Chelsea Finn described Mobile ALOHA’s two main functions.  Mobile ALOHA has a teleoperation system that makes it possible to collect demonstration data for complex tasks in an intuitive way, Finn explained. This allows the robot to perform tasks such as cooking shrimp after data collection rather than step by step manual programming. Finn furthered that the Mobile ALOHA proves that robots can easily learn from data collected through teleoperation to autonomously complete such complex tasks.

Stanford PhD student Zipeng Fu, the co-lead for project Mobile ALOHA, explains the challenges in the 3 month development process of the robot. He describes the technical challenges to be two-fold, hardware challenges and software challenges. In the past, researchers have used expensive hardware and off-the-shelf manipulation robots bought from manufacturers to run tests. The Mobile ALOHA team battled this common practice by assembling their own hardware, and designing a high-quality and low-cost teleoperation system. 

Another software challenge that presented itself was the need for step by step, explicit manual programming for the successful execution of daily tasks by robots. Team Mobile ALOHA decided to take a different route than the typical manual programming.

“We [took] a data-driven AI approach using human demonstration data to teach the robot (i.e. imitation learning). We showed that imitation learning, together with a co-training technique, is effective to teach the robot new autonomous skills using [less than] 50 demonstrations,” Fu said.

Ultimately, Mobile ALOHA made a breakthrough in the field of AI and technology. Mobile ALOHA shows the efficiency of imitation learning for teaching robots new skills and the importance of high-quality data collection. The project further proves that the simplification of robot programming and performance of daily tasks is possible. 

The Mobile ALOHA team hopes that their project and demonstration will spark further technological breakthroughs.

“Our long-term goal is enabling intelligent robots that can help human society,” Fu said. “We open-sourced this project with a hope of promoting research in the real-world robotics area for the public good, and attracting more people to work on [autonomous] home robots.”