Stanford and Google DeepMind researchers have presented an open-source housekeeping robot, and trained it relatively quickly to sauté shrimp, rinse out pans, put pots away in a kitchen cabinet, and clean up wine spills – but it has greater ambitions.
The Mobile ALOHA platform is far from a sexy-looking machine; there's a flattish wheeled base with a 12-hour battery pack weighing it down, supporting a brutally ugly mess of scaffolding, a laptop, a top camera and a pair of clawed robotic arms with wrist cameras on each, and 14 degrees of freedom.
Onto this aesthetic disaster, a removable training setup can then be grafted, which gives an operator the ability to push the bot around, and operate its arms and claws manually to train it.
This 75 kg (165-lb) robot is completely open source; the project team gives you a parts list and a guide on how to build the thing for yourself. It's a relatively low-cost build considering its capabilities – but that doesn't make it cheap; you can get the parts for a little over US$30,000 if you're handy with a glue gun and a 3D printer.
You can then train away, augmenting skills the ALOHA system already knows from previous training. The team says running through a task 50 times increases the chance the robot can do a task autonomously by up to 90%, and it's released video showing some of the hands-off skills it's picked up.
But the platform is capable of so much more, as this second video demonstrates. None of this footage is autonomous – the bot is being tele-operated. But it gives an insight into the kinds of things we can expect housekeeping robots to take over in the not too distant future.
And for a dash of realism, here's a third video showing why Mobile ALOHA is still a long way from a commercial product: it still makes plenty of dumbass mistakes, so you wouldn't want it to be handling the fine china just yet.
It may be just a student-led research project, but it's part of a growing wave of mobile robots demonstrating an accelerating ability to watch, learn and autonomously repeat tasks in a range of dynamic real-world situations.
It suddenly seems possible that the end of chores might not be far around the corner ... I can't say I'm gonna miss those.
Source: Mobile ALOHA