Doctors Discuss the Development of Robotic Hands (Part 2)
The intelligence and reasoning used to manipulate items with the human hands is something most people simply take for granted. To pick brush our hair, we effortlessly reach out and grab the brush and stroke it through our hair. When consuming food, we use utensils to precisely grasp our food and smoothly move it into our mouths. Even small children can easily accomplish these tasks without much conscious thought.
But getting a robot to this same level of intelligence will require a leap in the current artificial intelligence or machine learning that robotic scientists use to control robotic hands.
Artificial Intelligence & Robotic Hands
Typically, to date, the software behind robotic manipulation has involved building complex and detailed mathematical models of the situation(s) with which the robot will be faced. The robot then uses that exact model to plan and execute its motion – or sequential motions.
Take for example the recent example of a robot whose task was to assemble an Ikea chair. Scientists created a software model that could recognize each individual bolt, screw and piece of metal – and that understood how each piece fit together with the others.
After being programmed, the robot could assemble the chair in a rapid 20 minutes. But, when the same robot was tasked with assembling a different Ikea product, it had absolutely no idea what to do.
Humans, on the other hand, develop manual manipulation skills very differently. Humans do not learn each task individually in a vacuum. Instead, humans absorb knowledge through example, practice, and trial and error. And then we can apply and extrapolate learned skills across new and different tasks both similar and different in nature. For example, if a person learns how to peel and chop a potato, they don’t have to start from scratch when chopping a carrot for the first time.
Machine Learning & Robotic Hands
The answer to teaching robots to also be able to do this likely lies in “machine learning.” This refers to the process that allows robots to understand which maneuvers are successful and which manipulations do not work – and then enables the robot to use that information in new situations that it has not encountered before.
In other words, machine learning allows the programmer to supply the robot with several examples of items that have been “annotated” with things such as: “here is good place to grab the object.’” The robot could then theoretically rely on past data to examine a completely new object and make an educated guess as to how to grasp it.
Currently, however, these type of machine-learning algorithms require that vast amounts of data about possible outcomes are entered before the robot can start to determine the best plan of action. And even then, the robot may need hundreds -or thousands- of attempts to maneuver an object before they determine what works.
So, while the mechanical technology may exist to “build” a human like hand, it will likely still be quite a while before machine learning methods are sophisticated enough to enable robots to learn entirely on their own, and interact with objects as quickly and seamlessly as humans can.
Detroit Area Hand Doctor
Despite the slow and painstaking progress that robotic scientists are making in the field of robot hand-control systems, the “autonomous manipulation” that humans take for granted is still many years away.
This should give everyone pause to appreciate and protect our own amazing hands and fingers. If you are suffering from any injury or pain in your fingers, wrist, elbow or arm, contact board certified Detroit area hand surgeon Doctor Rehman for a comprehensive evaluation and consultation. As with any medical conditions, early detection, awareness, and a prevention or treatment plan is the most effective way to combat hand injury – and ensure a complete and rapid recovery.
Doctor Rehman will assess your individual situation, and prescribe the treatments that are best for your condition.