A team of researchers from the Massachusetts Institute of Technology and Stanford University is developing a new AI technique to help robots learn tasks such as packing or storing items in the trunk of a car.
Packing a car trunk for a road trip may seem like a simple task, but it has never been easy for robots to learn. However, a new study has put the training of these tasks in the hands of artificial intelligence .
Now, researchers at the Massachusetts Institute of Technology (MIT) and Stanford University have developed a new AI technique to teach robots to pack objects into a limited space – such as the trunk of a car or a suitcase – by overcoming a number of difficulties.
Specifically, this project has demonstrated how artificial intelligence (AI) can revolutionize a tedious task, taking the packaging capabilities of robots to a whole new level. Although this work has not yet passed peer review (i.e., peer review or evaluation by other scientists), according to experts, this innovation could have a future in different aspects: from domestic organization to the colonization of Mars .
During the experiment,
MIT researchers employed a type of AI known as a ‘diffusion model’ to train robots in the ‘art’ of packing. This technique has allowed robots to learn how to store items in limited spaces while overcoming a number of difficulties, such as making sure that heavier items do not crush lighter ones or preventing the robot arm from hitting the luggage and damaging it.
“We want to have a method based on machine learning to solve some limitations quickly, because this way AI will solve problems faster than traditional methods,” Zhutian ‘Skye’ Yang , lead author of the paper and a graduate student, explained to Scientific American.
Specifically, he explains that his method is based on ‘ BRD learning ‘, which consists of allowing an AI program to learn autonomously by identifying patterns between the training data and the desired result.
“Packaging with robots is incredibly difficult, but revolutionary,” the researchers say. “This work allows robots to start ‘thinking’ on the fly and quickly reach very good, if not optimal, solutions,” they add.
But the highlight of this approach has been its ability to quickly solve multiple storage challenges simultaneously . While traditional methods required robots to try out different ways of packing and verify each one, this model has allowed robots to explore a variety of machine learning models at the same time.
This new option has given them a more complete view of the problem and the ability to consider all packaging obstacles almost simultaneously. “There can be many solutions that are not intuitive. And you can change the plan on the fly,” Yang says.
Artificial intelligence for multiple tasks
For the authors of this feat, the impact of their study goes beyond simply packing for a road trip or storing groceries in the trunk. This technology is expected to have applications in a wide range of industries, from logistics and transportation to manufacturing or space exploration .
For example, in the transportation industry , robots could help shipping companies more efficiently pack a variety of items in a single space, which could reduce costs and improve delivery speed.
In manufacturing , machines could potentially streamline the packaging of products on assembly lines, increasing productivity and reducing waste. Even in space exploration , robots could be instrumental in storing resources and supplies for potential missions to places like Mars .
In addition to its impact on ability and productivity, this technology also has the potential to change the way we pack. Humans often follow certain instinctive guidelines when packing a chest, such as placing heavy items at the bottom or assign them evenly in a space. However, robots are not limited by these human conventions and could discover new solutions that might go unnoticed by us.
As this technology continues to develop,
there may be some issues that need to be considered, such as security or privacy , experts say. For example, how to ensure that robots do not damage items in the process or how to protect the sensitive information that these devices may collect about our habits and consumption.
Despite these challenges, AI’s potential to improve this task is undeniable. Over time, this technology is expected to transform the way we pack, from our suitcases to cargo containers on shipping vessels. “I want to have robots in the kitchen helping with household chores,” Yang says.
And the possibilities extend beyond the borders of our planet. “If you go to Mars, you can have a robot decide how best to package resources,” Yang suggests. Now, the team at MIT and Stanford University is working to make their robots even more capable of making “point decisions.”
This involves not only teaching a robot to pack within certain limits, but also training it to do so while considering continuously changing variables – for example, by having it put away objects while simultaneously moving around a room.
Read also: In this village, all the houses have been 3D printed
