If you have ever wanted a robot to cook food for you, the day is not far away. Stanford University has introduced Mobile ALOHA, an innovative system for learning bimanual mobile manipulation through low-cost whole-body teleoperation. This robotics technology addresses the limitations of traditional imitation learning from human demonstrations, which often focus on tabletop manipulation without the necessary mobility and dexterity for real-world applications.
Introducing 𝐌𝐨𝐛𝐢𝐥𝐞 𝐀𝐋𝐎𝐇𝐀🏄 — Hardware! A low-cost, open-source, mobile manipulator. One of the most high-effort projects in my past 5yrs! Not possible without co-lead @zipengfu and @chelseabfinn. At the end, what's better than cooking yourself a meal with the 🤖🧑🍳 pic.twitter.com/iNBIY1tkcB
Provided by Google Deepmind, Mobile ALOHA expands upon the existing ALOHA system by incorporating a mobile base and a whole-body teleoperation interface, making it capable of imitating complex mobile manipulation tasks.
The system’s primary purpose is data collection, allowing it to learn and replicate various bimanual activities. This includes 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.
Its ability to co-train with existing static ALOHA datasets sets Mobile ALOHA apart, significantly enhancing its performance on mobile manipulation tasks.
The research team also found that with just 50 demonstrations for each task, co-training can boost success rates by up to 90%.
This remarkable improvement allows Mobile ALOHA to handle complex and dynamic scenarios, showcasing its potential for real-world applications beyond traditional robotics limitations. From enhancing efficiency in kitchen tasks to navigating complex environments such as elevators, this breakthrough opens doors for a new era in robotics, where machines can perform a wide range of mobile manipulation tasks with precision and adaptability.
One of the key features of Mobile ALOHA is its cost-effectiveness, making it an accessible and practical solution for advancing robotics research. The system leverages supervised behaviour cloning, using data collected during teleoperation to train the robot to perform tasks autonomously.
2023 witnessed notable advancements in the field of robotics. For instance, Boston Dynamics upgraded Atlas for intricate construction tasks, and Microsoft empowered ChatGPT to command a robotic arm and drone.
Elon Musk’s Tesla is working on developments for its humanoid robot, Optimus. Demonstrations in September about performing Yoga and in December 2023 handling sensitive objects such as eggs showcased Optimus’ evolving capabilities.
Built Robotics introduced the RPD 35 solar piling robot, and robotics ventures received a $1.63 billion influx in April. Viam software became generally available, supporting robotic developers, while a collaboration produced a ChatGPT-driven tomato-picking robot.
