As a Select Voice of Google DeepMind’s AI Community, we are delighted to announce their release of research into robot dexterity, a key research challenge in developing more useful robots
In March 2024, Teens in AI Founder & CEO Elena Sinel was invited as part of a select group of voices from the AI community to a private, off-the-record meeting hosted by Google DeepMind. One of the systems they were introduced to as part of their research was DemoStart.
We are delighted to share Google DeepMind’s announcement and launch of two new AI systems that help robots learn and perform complex tasks that require a high level of dexterity. We are even more excited that this comes at a time when, in early 2025, Teens in AI will be releasing two new courses on ROBOTICS (watch this space)!
Learning to tie shoelaces may come easily for people but it’s incredibly difficult for robots to learn, as this level of dexterity requires very precise, fast movements, and a dynamic understanding of the environment and object.
Solving robot dexterity is key to developing useful robots that can assist with many everyday tasks. The Google DeepMind robotics team has two new projects that tackle this challenge:
🤖 ALOHA Unleashed method enabled robots to achieve five highly complex and novel bi-arm manipulation tasks: tie a shoelace, hang a shirt, repair another robot, insert a gear, and clean a kitchen. It applies a diffusion method, predicting robot actions from random noise, similar to how our Imagen model generates images.
🤖 DemoStart uses reinforcement learning to help robots learn dexterous behaviors from a few demonstrations in simulation and perform in the real world. It requires 100x fewer simulated demonstrations to learn how to solve a task in simulation – like tightening a nut-and-bolt – than what’s usually needed when learning from real world examples for the same purpose.
These AI systems are still a long way from achieving human-level dexterity, but represent an important step towards AI robots that could help people with all kinds of tasks at home, in the workplace and more.
Google DeepMind's latest advances in robot dexterity
Two new AI systems, ALOHA Unleashed and DemoStart, help robots learn to perform complex tasks that require dexterous movement. Until now, most advanced AI robots have only been able to pick up and place objects using a single arm. In their new research paper, they present ALOHA Unleashed, which achieves a high level of dexterity in bi-arm manipulation. With this new method, our robot learned to tie a shoelace, hang a shirt, repair another robot, insert a gear and even clean a kitchen.
