• Google's RT-2: A visionary AI robot with superhero-like abilities - perceiving, reasoning, and acting like a pro!
  • RT-2 breaks the clumsy robot communication game, impressively transferring knowledge among robots.
  • Monkeying around in the tech world, RT-2 excels in novel tasks, learning like a human with fantastic success rates

Robots and AI are truly becoming the guardians of the galaxy. I sometimes get overwhelmed with the imagination of robots taking over us and AI corrupting our minds, just like in the series ‘Black Mirror’ on Netflix. But wait, what if they don't actually turn out to corrupt our minds but instead become the monkeys of the tech world? 

When was it released?

It was on the 28th of July 2023, Google launched RT-2 (Robotic Transformer -2) which is a first-of-its-kind vision-language-action (VLA) model.

What is a VLA Model? 

The VLA model is like giving them superpowers! VLA stands for "Perceive, Reason, Act" – three crucial steps in their heroic adventures.

First, they perceive the world around them using sensors, just like your eyes and ears. They're like robot spies, gathering info on what's happening.

Next comes reason, where their brains analyze all that data like a genius detective figuring out the best plan. It's like they're thinking, "Hmm, if I see a wall, I better turn left."

Finally, the grand finale is the act! They execute their brilliant plan – moving, grabbing stuff, or even dancing! It's like a superhero pulling off an epic move!

What is special about it?

Google has developed this technology but with a new approach. Thanks to developments like chain-of-thought prompting and complex vision models like PaLM-E, robots have made progress in reasoning and perception.

RT-1 demonstrated that robots of different types could communicate and share knowledge. But let's face it, controlling robots has been a clumsy telephone game up until now. Imagine that robots had to communicate their intents through a maze of intricate networks. It's similar to wanting to dance but having to communicate each motion to your limbs individually. What an ineffective system!

Here comes RT-2, the robotics industry's game-changer. This bad boy creates a single coherent model that combines the strength of sophisticated reasoning with fluid action.

There will be no more sporadic communication or acting as a mediator between the brain and body. RT-2 exhibits true multitasking capability as she moves and thinks with effortless ease. The RT-2's capacity to transfer knowledge amongst robots, even those doing diverse jobs, is simply astounding.

In a manner similar to a robot masterclass, they impart their knowledge and abilities to one another. The best thing, though? This is accomplished using RT-2 with little training data. Highest level of effectiveness! But hey, I feel, it can be a more smarter since Google is training it as some chatbot. 

According to Google, in testing RT-2 models in more than 6,000 robotic trials, the team found that RT-2 functioned as well as our previous model, RT-1, on tasks in its training data, or “seen” tasks. And it almost doubled its performance on novel, unseen scenarios to 62% from RT-1’s 32%.

In other words, RT-2 enables robots to learn more similarly to humans by applying previously learned concepts to fresh contexts. Yes, Monkeys of Tech World. 

Although Google’s Bard AI might have updated to languages and increased its reach after a very long time, hopefully, RT-2 will get elevated and upgraded soon with lesser complexities and more success rate.  


Are you working on an innovative product? We would love to hear your story. Share it with us at [email protected]

Edited by Shruti Thapa