Google DeepMind Taught a Robot to Tie Shoes Very Poorly
Google DeepMind has announced that it has taught one of its robots how to tie a shoe. Unfortunately, the robot isn’t very good at it yet. In fact, the bot’s shoe-tying abilities arguably rank quite low on the overall scale. I’m pretty sure most American adults (and also children) can tie their shoes better than this robot.
The bot’s breakthrough was announced in a tweet posted on X on Thursday. The tweet included a video of the machine’s technical feat:
Manipulating objects using just one robot arm is challenging.
Enter ALOHA Unleashed, which builds upon our ALOHA 2 system. With two arms, it can be teleoperated to collect high quality training data.
With this system, robots can perform new tasks with fewer demonstrations. pic.twitter.com/D6J0P3eqxJ
— Google DeepMind (@GoogleDeepMind) September 12, 2024
As you can see from the video, those are some really big “bunny ears.” Just imagine trying to walk around in those things. You’d be face-planting immediately.
In a blog post entitled “Our latest advances in robot dexterity,” DeepMind goes into further detail about how they taught the machine to lace up. Unusually, the bot uses two arms to complete the task—or what researchers dub “bi-arm manipulation.” The bot is also powered by two new AI systems, dubbed ALOHA Unleashed and DemoStart. ALOHA’s programming helps the bot perform difficult, two-armed tasks, while DemoStart involves the use of simulations to help the robot learn from its experiences. “One day, AI robots will help people with all kinds of tasks at home, in the workplace and more,” researchers write. “Dexterity research, including the efficient and general learning approaches we’ve described today, will help make that future possible.”
Many companies are currently racing to build robots that can engage in human-like activities. Minor tasks like shoe-tying or shirt-folding have long been considered a critical litmus test of a robot’s abilities. Some companies are having more success than others. Earlier this year, Elon Musk shared a video of his Tesla bot (dubbed “Optimus”) folding a shirt. It later turned out that the video had been partially staged and that the robot could not actually fold a shirt autonomously yet.
The presumable end game for this type of technology is autonomous manufacturing plants, where companies can use robots to make products from scratch without having to hire any workers. It’s unclear why else you would build a machine whose purpose is to replicate labor already accomplished by humans. I guess we can all take comfort in the fact that the robots that exist today sure don’t seem like they’ll be up to the task anytime soon.