Google’s DeepMind organization seems to be cooking up a storm after it just made some breakthrough revelations. This was in terms of how it feels Robots can be safer and faster at what they do.
This entails a system for combining training data using Robot Constitution. The latter would transform assistants into doing tasks more safely and effectively. For instance, it would be trained to fetch papers for printing without bumping into a human colleague along the way.
Meanwhile, the AutoRT which is the company’s system for gathering information would use visual language models alongside the usual LLM to comprehend the kind of environment working around them.
It would adapt to unfamiliar settings and also decide on the right kind of tasks taking place.
The Robot Constituion is a unique set of laws that were designed based on inspiration gathered from the document by Issac Asimov who explained more safety principles and prompts on how to prevent selecting tasks involving humans and other appliances.
For more safety, the company designed the robots in a manner that automatically prevented forces from going above and beyond a listed threshold. This entailed physical switches that could turn the robot off when the human feels it might not working appropriately.
In just seven months, we’ve seen the tech giant rollout fleets featuring 53 robots in four different office locations while carrying out more than 77k to ensure things work smoothly with the robots in place.
Some of the robots were seen carrying out functions with human operators in charge and then some carried out tasks using scripts or through autonomous means.
But don’t be fooled by the looks as they are not trendy or flashy in design. However, they are equipped with great potential, featuring the likes of cameras, robot arms, and whatnot. Every system makes use of VLM to better gauge the environment surrounding them.
Meanwhile, LLM makes use of creative list suggestions that robots are asked to carry out. And while we might not be too near to the era where robots serving drinks, snacks, and pillows for our comfort autonomously arise, when they are rolled out, they’re bound to have received great training from the best systems including AutoRT.
Read next: Google Rolls Out First Test For Its Highly Anticipated Cookie-Removal Strategy
