Researchers at the Massachusetts Institute of Technology (MIT) have created a deep learning neural network that “teaches” robots to position themselves autonomously in space, detecting their location and also all parts of their body. These conditions increase the efficiency of robotic devices and diversify their potential applications.
According to a release, the new features provided by the Artificial Intelligence system make it possible for soft-body robots to better interact with their environment and complete the tasks assigned in each case. The breakthrough is a step toward automating robot design, as the neural network is able to figure out how to best design each robot to solve a specific task.
The MIT scientists set out in their study, published in the IEEE Robotics and Automation Letters journal, the realization of a technical advance that allows white-body robots to better relate to their environment and their own structure. It is that unlike rigid leather devices, these robots can vary between a practically infinite number of forms.
Although this condition increases their versatility and makes them more friendly to humans, at the same time they make it difficult to position them in space and to detect each part of their structure. Consequently, these robots often fail to relate adequately to their environment and lose possibilities in terms of efficiency and resolution capacity.
Solving location problems in space
While rigid robots have a limited range of motion, due to the presence of a finite set of joints and limbs that allow for manageable calculations for motion mapping and planning, the situation for soft robots is very different. It is that any point of a soft-bodied robot can, in theory, be deformed in all possible ways. Logically, this multiplies the possibilities of movement and makes its design very complex.
Now, the algorithm created by American researchers is able to help engineers design soft robots that collect more useful information about their environment. Integrated into a deep learning system, the algorithm suggests an optimized location of sensors within the body of the robot, so that in this way it better captures the environment and can successfully perform new tasks.
Thanks to this innovation, specialists believe they have taken an important step towards the development of truly efficient white body robots. It is worth noting that the neural network incorporates the characteristics of each part of the robot’s body, to later carry out a trial and error process through which it “learns” the most efficient sequence of movements to complete a given task, such as grabbing. objects of different sizes.
To verify the efficiency of the system, the researchers compared the work of robots with sensors optimized by their algorithm with that of others that had sensors placed based on the criteria of experts. The results were conclusive: robots designed by the Artificial Intelligence system clearly outperformed robots optimized by humans in terms of efficiency and ability to solve tasks.
Based on these results, the scientists believe that this technological solution could open a new path towards intelligent automation of the design of robots, increasing its impact in various areas of industrial and economic activity, as well as its use in assistance tasks and collaboration with the human being.
Co-Learning of Task and Sensor Placement for Soft Robotics. Andrew Spielberg; Alexander Amini; Lillian Chin; Wojciech Matusik; Daniela Rus et al. IEEE Robotics and Automation Letters (2021).DOI:https://doi.org/10.1109/LRA.2021.3056369
Video: Alexander Amini.
Photo: Xu Haiwei en Unsplash.
Eddie is an Australian news reporter with over 9 years in the industry and has published on Forbes and tech crunch.