Mobile Aloha | More Like Mobile Hallelujah
Advancements in humanoid robotic technology have been focused on assisting humans in various non-productive tasks. With this mission in mind, a team of researchers from Stanford University, Google DeepMind, Berkeley University, and Meta has collaborated to create Mobile Aloha, a potential AI-powered housemaid of the future. This new advancement builds upon Google DeepMind’s Aloha system and emphasizes the importance of mobility and agility in the field of robotic learning.
Advanced Proposition: Mobile Aloha
Mobile Aloha is a humanoid system that integrates a mobile base and a comprehensive teleoperation interface for whole-body control. Unlike conventional imitation learning centered around tabletop scenarios, Mobile Aloha can emulate intricate mobile manipulation tasks. It serves the purpose of data collection, enabling the acquisition of knowledge and the reproduction of diverse bimanual activities.
One of the key features of Mobile Aloha is its capability to simultaneously train with the established static Aloha datasets, setting it apart from regular robotic systems. Through a combination of supervised behavior cloning and 50 demonstrations for each task, researchers claim that Mobile Aloha enhances its performance in mobile manipulation tasks by up to 90 percent.
Mobile Aloha’s breakthrough enables the robot to independently navigate through complex situations, such as cooking shrimp, organizing household items, opening cabinets to store heavy pots, calling and entering elevators, and rinsing pans using a kitchen faucet.
Mobile Aloha builds upon the advantages of the original Aloha system, featuring a cost-effective, dexterous, and repairable bimanual teleoperation setup. It is designed to handle substantial household objects like pots and cabinets while maintaining stability. The system allows for whole-body teleoperation, enabling simultaneous control of all degrees of freedom, including both arms and the mobile base.
Additionally, Mobile Aloha is untethered and equipped with onboard power and computing capabilities. It incorporates a 1.26 kilowatt per hour (kWh) battery weighing 30.8 pounds (14 kilograms) to provide power and act as a counterbalance, preventing the system from tipping over. All computing tasks during data collection and decision-making are performed on a regular laptop with a Nvidia 3070 Ti GPU (8GB VRAM) and an Intel i7-12800H processor.
The system utilizes three Logitech C922x RGB webcams, with two attached to the wrists of the follower robots and one facing forward. These webcams provide video streams to the laptop, which also collects information about the arms and the mobile base. This data allows the team to capture the linear and angular speeds of the mobile base and record the joint positions of all four robot arms, serving as observations and actions for the policy.
Capabilities and Cost
Mobile Aloha possesses a vertical reach from 65 to 20 centimeters (cm) above the ground and an extension capability of 100 cm beyond its base. It can lift objects weighing up to 3.3 pounds (1.5 kilograms) and exert a pulling force of 100 newtons (N) at a height of 1.5 meters (m). The whole system, costing around $32,000, is expected to become more affordable if it enters mass production.
Mobile Aloha represents a significant advancement in humanoid robotic technology, specifically in the field of AI-powered housekeeping. With its mobility, agility, and comprehensive teleoperation interface, Mobile Aloha can perform complex tasks, such as cooking and cleaning, with impressive success rates. This innovation opens up possibilities for future applications in various industries, where humanoid systems can assist humans in non-productive tasks, ultimately enhancing efficiency and productivity.