ChatGPT is tested on robots – the chatbot successfully controls drones and distributes tasks to robotic assistants.

ChatGPT has demonstrated to the world its ability to quickly answer all kinds of questions, write posts for social networks, articles and even books. Now they are trying to teach a chatbot to control robotics: for example, a home robot assistant or even a drone.

Microsoft quickly responded to the popularity of ChatGPT – it is now successfully integrated into the Bing search engine, and will soon appear in the Edge browser and the corporation’s office products. Microsoft is rumored to have spent more than $10 billion investing in chatbot developer OpenAI startup, so it plans to get as much out of ChatGPT as possible.

Currently, the company’s researchers are testing ChatGPT on robotic equipment and visually demonstrating how AI can control a mechanical arm or even a drone. The team says it’s looking into ways to improve human interaction with the robots so they can respond to our commands using the code that ChatGPT has created for them.

“Well, wouldn’t it be great to say to a robotic assistant, ‘Please warm up my lunch,’ and it would find a microwave oven on its own?” Although language is the most intuitive form of expressing our intentions, we rely on handwritten code to control robots. We will try to change this with ChatGPT”, – they write on the Microsoft page.

ChatGPT is a language model trained on a large amount of text and human interactions, which allows it to generate realistic answers to any question (although not always true). The key task of the new research is to teach the chatbot to solve tasks based on the user’s request, the laws of physics and the context of the robot’s operating environment.

During one phase of the experiment, the chatbot was given access to functions that could be used to control a real drone, and it turned out to be an extremely useful interface that connected the user with the robot. For example, ChatGPT asked clarifying questions when the instructions were not very clear, and successfully wrote complex code structures for the drone – for example, asking it to perform a zigzag flight to visually inspect the shelves. Oh, and he learned how to take a selfie.

ChatGPT also skillfully manipulated the robotic arm by writing the code to assemble the blocks. Later, the chatbot combined its textual and physical skills to create the Microsoft logo—it was able to recall what the logo looked like by consulting its own database, draw it as SVG code, and then analyze what skills the robot would use to form the required shape and give it appropriate command.

Source: Microsoft

Related Posts

Nokia updated its logo for the first time in 60 years

The once-legendary Finnish smartphone maker Nokia Corporation has unveiled a new logo — its first logo change in nearly 60 years. The brand’s new identity, which causes…

“Not many people know what Nike is.” Ukrainian trailer of the film “AIR” / Air about the success story of a sports brand with Ben Affleck and Matt Damon

Warner Bros. Ukraine has published a dubbed Ukrainian trailer for the new film directed by Ben Affleck called Air, about the success story of the world-famous sportswear…

Review of the film “Myrnyi-21”.

Russia’s military invasion of Ukraine in 2014 forever changed the lives of all Ukrainians. This was reflected in all spheres of practical life in the country. Cinematography…

We have a ghost

On February 24, Netflix released the comedy film “We’ve Got a Ghost!”, starring David Harbor and Anthony Mackie. Its plot is based on the short story “Ernest”…

Microsoft is pushing users to install Windows 11 even on non-supported PCs with full-screen banners.

Microsoft Corporation is very active in luring Windows 10 users to the more modern Windows 11. This is far from news. But now the company’s efforts in…

A neural network in a smartphone. Qualcomm optimized Stable Diffusion for fast performance on Snapdragon 8 Gen 2

Bringing large language models like Stable Diffusion to smartphones is the next step in bringing generative tools to a wider audience. Typically, such generative neural networks require…

Leave a Reply

Your email address will not be published. Required fields are marked *