Meta-AI that teaches robots how to organize rooms
When we talk about artificial intelligence, we usually immediately think of ChatGPT since it is the most popular in this field. A chatbot is capable of doing a lot of work for us if we give it orders, which it can carry out through text messages. However, AI is not limited to only this as it has also started to be used in devices. Meta has developed an artificial intelligence system called OK-Robotwhich is responsible for training a wide variety of robots tidy up the room by collecting items.
As technology advances, we have life with great amenities which means we have to work less and less on our own. For example, thanks to the Internet, we can buy things, make reservations, or even perform treatments right from home. mobile, Being in sofa or in bed. Instead of going to physical sites to buy groceries or complete paperwork, we do most of it from our mobile phone or computer. The same can be said about connecting a TV or light: for this we can use voice.
OK-Robot is an artificial intelligence system that allows robots to identify and pick up objects.
We can use Google Assistant, Siri or Alexa to be able to search for videos on YouTube or anywhere else on the Internet, and also to be able to turn off smart lights (if we have a home prepared for this). All this allows us to take small steps towards that futuristic society that we have seen so many times in films and TV series. However, something fundamental is missing – human robots or androids living next to us. We haven’t reached that point yet, but we can say that robots are capable of doing more and more things.
Meta showed what his OK-Robot AI is capable of, which uses VLM able to process and understand text and images. This uses several open source artificial intelligence models and was pre-trained on large data sets. This has led to the emergence of AI, which can be embedded in a robot and operated directly. without training.
After conducting 171 experiments, a success rate of 82.2% was recorded.
To see the results, OK-Robot’s AI was tested using Hello Robot’s commercial Stretch robot. In total this amounted to 171 experiments on pickup and delivery parcels to different houses. During these tests, the robot scanned its environment using an app. iPhone Recording3D and created a 3D video. It then applied an AI-powered object recognition model to each frame of said video, allowing it to identify anything that was useful.
Once you saw the items you needed, you were ordered to collect them. In this case, it was observed that there were success in 82.2% of cases. We are talking about tests in a slightly cluttered room, since the more chaotic be, the less chance of success you will have when ordering it. The downside is that the robot can only pick up one object at a time and leave it somewhere else, so in the future it will be able to perform more than one task at a time.