Imitation learning has shown great potential in robotics, especially in table-top manipulation. A team of researchers from Stanford University and Google DeepMind has made progress in this field by creating a Mobile ALOHA (A Low-cost Open-source Hardware System for Bimanual Teleoperation) humanoid system.
This robot can perform multiple activities like cooking shrimp or organizing things in your house, making it a potential AI-powered housemaid of the future.
Building on Google DeepMind’s ALOHA system, this novel contribution addresses the high costs and technical challenges of training mobile bimanual robots that require careful guidance from human operators. The project also involved researchers from Berkeley University and Meta.
It costs a fraction of off-the-shelf systems and can learn from as few as 50 human demonstrations. This system came out at a time when there was a significant acceleration in robotics, and the success of generative models partly enabled it.
The Mobile ALOHA integrates a mobile base and a comprehensive teleoperation interface to enable whole-body control. The system is designed to emulate intricate mobile manipulation tasks, which were previously challenging using conventional imitation learning techniques focused on tabletop scenarios. ‘
Mobile ALOHA is primarily used for data collection, laying the groundwork for acquiring knowledge and reproducing diverse bimanual activities.
One of the key features of Mobile ALOHA is its ability to simultaneously train with the established static ALOHA datasets, which sets it apart from regular robotic systems. The system achieves impressive success rates using a combination of supervised behavior cloning and 50 demonstrations for each task.
As per the researchers, this enhances its performance in mobile manipulation tasks by up to 90%, allowing Mobile ALOHA to autonomously complete complex mobile manipulation tasks such as sauteing 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.
While the robotic chef does impress with its abilities, there are still some outtakes where it bumbles and gets things wrong. Nonetheless, this is a significant development in robotic learning from instruction. It remains to be seen how AI will impact human jobs, but the current administration has taken steps to draft governance and guidance around the technology. The hope is to find a healthy balance that safeguards current jobs while addressing the potential for both promise and peril that artificial intelligence holds.
