Using a New Artificial Intelligence Algorithm, MIT has designed Soft Robots That can Sense
Though traditional robots are significant investments, they still aren’t made for everything, especially the rigid and metallic ones. Where, if we talk about soft-bodied robots, they can interact as well as can seamlessly slip into tight spaces.
However, robots rely entirely on programming to perform their duties, and they have to know about every information of their body parts. It would be a lengthy task for those soft robots that can disfigure in infinite ways, but virtually.
What the researchers of MIT developed
They have developed an algorithm that would help engineers develop soft robots that can collect information about their surroundings more usefully. With the help of a deep-learning algorithm placed as a sensor within the soft robot’s body, it allows it to interact with the environment and complete tasks efficiently.
Alexander Amini, one of the researchers, said that the system could learn the assigned task and found the best ways to complete the job. He added that the sensor placement was difficult, but it would be fascinating to like the solution.
This research on Soft Robotics would be present during this year’s April IEEE International Conference and would be publishing the IEEE Robotics and Automation Letters journal. And guess what? The co-lead authors would be Alexander Amini and Andrew Spielberg, two Ph.D. at Computer Science and Artificial Intelligence Laboratory, MIT. And the other co-authors would be Lillian Chin, Ph.D. student (MIT), and Prof. Wojciech Matusik, and Daniela Rus.
Also Read: Artificial Intelligence Reshapes Management
Why the research was done on Soft-bodied robots
Still, it’s a big challenge for engineers to make soft-bodied robots that would complete real-world tasks. Because their counterparts have a built-in advantage and rigid counterparts, there is a limited range of motion.
In contrast to soft robots, rigid ones have a finite array of limbs and joints, making it possible to perform algorithmic calculations that would control the motion and mapping. According to Spielberg, the problem with soft-bodied robots is that they have infinite dimensions and can deform in any way, as per theory.
Yes, that’s a disadvantage because it makes designing the soft robot and mapping the location of its body parts difficult. In the past, there have been efforts on charting the robot’s position using an external camera and feed the collected information back into the robot. However, the researcher at MIT wanted to create a soft robot without any external devices or cameras.
More about the research to design AI-powered Soft Robot
Spielberg said that, on the soft robot, you could put an infinite number of sensors, and with that, arise the question of the number of sensors they would need to set. With that, the team turned towards deep learning to get an answer to the question.
The researchers then developed an architecture made of a novel neural network that would optimize the sensor placement and complete tasks by learning efficiently. At first, the researchers divided the soft robot’s body into particles, that is, regions. Through trial and error, the network would learn the most efficient sequence of movements to complete any task.
Also, the network keeps an eye on the particles or regions that are primarily used and culls the lesser-used particles for the network’s subsequent trials from the set of inputs. It optimizes the important particles with the network, suggesting where the sensor should ensure maximum performance.
For example, the grasping hand of a simulated robot would have an algorithm that would suggest the sensors get concentrated in and around the fingers, precisely controlling the interactions with the environment.
That shows that the algorithms could easily outperform human intuitions when finding the site for the sensor. The researchers set the algorithm against the series of expert predictions.
There were four different layouts of soft robots; the team asked roboticists to manually select the sensor placement that would enable efficient completion of tasks. After that, they ran simulations to compare algorithm-sensorized robots and human-sensorized robots.
The results were not close; Amini said that their model outperformed humans in a vast way for every task; he looked at some robot bodies and could figure out where the sensor would fit.
Conclusion
Spielberg said that their research would help automate the process of robot design. He added that other than developing algorithms to control the robot’s movement, they would need to think about sensorize the robot that will interplay with the rest of the system’s components.
After the placement of the sensor in the correct way would help in industrial applications, where these robots would be used for the fine task. According to Spielberg, there is potential for immediate impact on the industries because that’s where they would need a robust, well-optimized sense of touch.
By automating the soft robot’s sensorized design, would be an essential step towards the rapid creation of intelligent tools helping people with physical tasks, said Rus. She added the sensors are a necessary factor of the process; they would enable soft robots to see and understand the world’s relationship and the world itself.
About The Author
