Improving Electric Prostheses with Ai

Prosthetics is the branch of surgery that “involves the use of artificial limbs to enhance the function and lifestyle ofpersons with limb loss” (“What Is Prosthetics”). These artificial limbs are called prostheses (singular: prosthesis). Many people include devices that replace parts of the body that are not limbs, such as glass eyes or pacemakers, in the definition of a prosthesis. This paper, however, will focus only on the replacement of limbs.

A prosthetic can be controlled in one of two ways: body-powered or electric. A body-powered limb is one that is totally manual, usually “[relying] on a system of cables or harnesses (along with manual controls, in many cases) to control the limb itself” (“Electric vs. Body-Powered”). Body-powered prosthesis are generally more affordable and reliable than electric ones. An electric limb, sometimes called myoelectric (myo meaning muscle), “[works] by using your existing muscles in [the] residual limb to control the functions of the prosthetic device itself” (“Electric vs. Body-Powered”). This results in more natural movements and finer motor control in the limb.

In the past, prostheses were bare-minimum substitutes for a missing limb, such as a metal rod that is connected to the remaining leg by a harness. Slowly, sticks and harnesses turned into elaborate designs that imitate the real limb. Functionality was added so that the amputee can grab objects or bend their knee with the help of manual controls and cables. Later, we developed artificial limbs that connect to the muscles in the remaining limb that control the limb electronically. Today, we have prostheses that can be roughly controlled using signals from the brain. These vast improvements are promising when you imagine the ultimate goal of prosthetics: to create artificial limbs that function exactly as well and as easily as a real limb. Current Efforts and the MediaÖssur, an Icelandic company that develops prostheses, seems to be at the forefront of research on AI-controlled prostheses. Their most notable design, the Rheo Knee 3, is claimed to learn a user’s gait in under 15 seconds. After minimal training and practice, it can climb stairs naturally and reliably.

The Rheo Knee 3 is said to learn continuously, which means it can adapt to new situations and environments without needing to be explicitly trained for it (Viejo). The idea of a continuously learning prosthesis is discussed thoroughly in a TED Talk by Dr. Patrick Pilarski, who is the Canada Research Chair in Machine Intelligence for Rehabilitation at the University of Alberta and is the lead of the Amii Adaptive Prosthetics Program, which is centered around creating intelligent prostheses. In his talk, he stresses the significance of continuously learning prostheses, giving the example of somebody taking up cooking as a hobby. He explains that a continuously learning prosthetic arm will learn the new motions of chopping and stirring, making it easier for the user to cook. If the prosthesis was only given things to learn during its initial training, it would not be able to remember the chopping and stirring motions, leaving the control completely to the user (Pilarski).

In the media, there is confusion surrounding prostheses and the science behind the progress of artificially intelligent prostheses. Many of the errors are attributed to using buzzwords like “bionic”, “AI”, and “smart”. In general, many articles will use the word bionic to describe any prosthesis that is both electric and controlled by the brain, and they use it interchangeably with words like “cybernetic” and “smart”. While those terms and concept overlap with bionics, the articles do not use the word accurately.

To clarify, bionics is the study and practice of creating artificial systems that closely mimic the functions and abilities of the living things they are designed to replace. The field of bionics emcompasses much more than just prosthetics, since the goal is to observe and imitate the most efficient natural processes and functions (“Bionics”). In prosthetics, bionics can be used to simulate natural hand movement in bionic hands. Cybernetics, while similar, deals with the control systems in place in living creatures. In prosthetics, cybernetics can increase the functionality of a hand that is connected to nerves and muscles (“Cybernetics”). By using the communication and control systems already in place in humans, we can begin to create limbs with infinitely many motions, rather than simple pre-programmed motions, such as grabbing. Another common misunderstanding was surrounding the actual implementation of AI.

Even in scholarly articles, it can be difficult to find information on AI in prostheses since it is often not explicitly stated that machine learning is being used to train the prosthesis. In several cases, popular tech news sites would describe the prosthesis as “smart”, which ended up not being a good indicator for the use of AI. Some news sites even seemed to think that AI was being used even if the prosthesis was controlled only by brain impulses or muscle movement, without the prosthesis learning or adapting in any way. Natural Motion, Reliable Action Predictions, and MoreIn an ideal world, prosthetic limbs would function just as well, if not better than, healthy limbs. The goal of creating artificial limbs that rival real limbs is, while not impossible, very difficult to achieve. There are several important areas that would benefit from improvement, including natural motion, more accurate predictions, and cost.

Natural Motion. Natural motion is difficult to master. To give some perspective, an able-bodied person has the benefit of their body and brain working in harmony to produce smooth and natural movements; however, this still takes years of practice and fine-tuning. Even after “perfecting” natural motion, the human body is continually improving and adapting to new situations. On the other hand, a person with a prosthetic arm, for example, does not have all of the benefits of having their limb controlled by their brain. While some prostheses respond to electrical impulses from the brain, the variety of motion is often limited to a set of predetermined actions, such as grabbing or pinching. In addition to this, the user has not had their entire life up to that point to practice with that particular prosthesis.

Machine learning algorithms are greatly reducing the time it takes to learn how to properly use a new prosthesis by having the user train it to know their individual gait, walking speed, environment, and so on. A prosthesis that uses AI to learn the behavior of its user can greatly improve the quality of life of an amputee by making it easier to perform day-to-day tasks, such as turning door knobs or climbing stairs. Predicting Movements. Learning about how the user moves is essential to predicting their next movements. Proper predictions are important because if the prediction is wrong, it could cause the user harm. For example, if a prosthetic leg wrongly predicts that it is about to walk up stairs and begins lifting the leg higher, this can cause the user to lose balance and fall unexpectedly. Since the user relies on the prosthesis to perform the correct action, the risk of a wrong prediction needs to be low.

A lot of effort is going into the research (Zhang) on which types of faulty predictions are safer and more convenient to make versus which can cause serious injury or a major inconvenience. In general, the research shows that for a prosthetic leg, any mistake made while the foot is in the air is usually safe and minorly inconvenient at most, while a mistake made while the leg is bearing a load (the weight of the body) is often dangerous or majorly inconvenient. Less Effort. With AI predictions and electric limbs, amputees will use less effort while performing simple or repetitive tasks. Rather than having to swing a body-powered prosthetic to walk, the leg will “walk itself” by applying force to the ground and bend at the knee. This greatly reduces the strain on the amputated limb and allows the user to focus on things other than balancing while they walk. Better Balance. Balancing on a prosthetic leg can be challenging, especially for elderly amputees.

A prosthesis that uses AI to help detect changes in weight distribution can balance more easily and reliably with no special input from the user. This helps people walk properly and safely on uneven ground as well as stand without a balancing aid. While automatic balancing benefits everyone, it especially benefits people who are at a higher risk of falling, such as those with weakened muscles at the end of the limb, elderly people, those who travel on subways, and hikers. Cost and Accessibility. The cost of an electric prosthesis can be anywhere from $3,000 to $50,000. As a reference, the Ossur Rheo Knee 3 costs an estimated $45,000 without insurance. In the US, medical insurance will cover most of the cost of a prosthesis if it is deemed medically necessary. However, there are many ways of making more inexpensive prostheses that use AI. Many people and companies have begun 3D printing body-powered prosthetic arms at a minimal cost.

Very recently, Joseph Sirosh at Microsoft has developed a prosthetic arm that connects to the cloud and uses computer vision to recognize objects and grab them in the correct way (O’Reilly). Sirosh states that this prosthesis only costs a few hundred dollars without insurance. Many of these goals are being attempted by teaching machine learning algorithms with reinforcement learning in a simulated environment. Łukasz Kidziński at Stanford University has created a physiologically-based human model with a prosthetic leg in a simulator called OpenSim. This human model is a musculoskeletal model, meaning that it has contracting muscles and rigid bones that simulate the different stresses on a human leg while it moves. This is a huge improvement on the typical “stick man” model that is commonly used when teaching an AI to walk or run, which lacks muscle and results in an abnormal walk or run.

By including a prosthesis on the model, the AI can find a more applicable solution to walking and running. Kidziński’s model is available on crowdAI as open source, attached to a challenge called the AI for Prosthetics Challenge in which the goal is to create an AI that adapts to changes in speed, direction, and environmental conditions the fastest (Kidziński). While training AI in a simulation will not perfectly match an amputee’s needs, it creates a good start for learning to walk, run, and climb without the need to physically train it, and the AI should be allowed to continue learning about its user to perfect the functionality over time. Prosthetic Hands that See and “Feel”

In robotics, it is common to use computer vision to help control a robotic arm’s movements. Research has already been done in “object recognition, arm positioning, grasping estimation, and vision feedback control” (Martin). This concept, however, is new to prosthetics. Adapting this research to a prostheses is not challenging, considering the prosthesis is similar enough to a robotic arm. A team of students spread throughout universities in Florida and Louisiana have created a working prototype of an arm that detects and grabs objects with the help of “eye gaze” data. In essence, the user will look at an object, the arm will recognize that the object is within reach, and then the arm will move and grab the object, avoiding any obstacles. Their prototype was successful, although it is not ready for widespread use since the user needs to wear a helmet for eye-tracking and connect the arm to an external computer (Martin).

A team of students at Newcastle University have improved on this concept with a prosthetic hand that recognizes different objects and adjusts the grab strength accordingly and can accurately predict the strength needed to grab and hold an object it has never seen before (“Hand That Sees”).

Artificial “feeling” hands need to be surgically implanted. This is because electrodes need to be placed at as many nerve endings as possible to be able to stimulate nerves and provide feedback to the brain. Using these haptic feedback prostheses, one man, named Igor Spetic, can pull cherries off of their stems with a 93% success rate, compared to a 43% success rate using the same prosthesis with haptic feedback turned off. The significance of sensations of touch in prosthetic limbs is tremendous, since being able to restore “one of the most basic forms of human contact” is incredibly important to amputees. When amputees are asked by researchers at the DARPA HAPTIX program, “universally they say they want to hold a loved one’s hand and really feel it” (Tyler).

Problems for AthletesSpecial athletic prostheses, such as running blades, are popular among many athletic amputees. The use of these prostheses in competitions are controversial, since some see the artificial limb as an enhancement, while others see it as a handicap. This puts athletic amputees in the strange situation of having an advantage in a sport while also having a disability. After Oscar Pistorius from South Africa competed in the Olympics with a blade prosthesis on each leg, another Olympian, named Markus Rehm, was denied permission to compete after failing to prove that his prosthesis did not give him an advantage. This led to studies on the topic, which resulted in the conclusion that a runner using bladed prostheses would use 17% less energy to run than an able-bodied competitor, while also taking 21% less time to swing the leg forward while running. These results led to bans on Olympic athletes with these types of prostheses (Greenemeier).

As AI-powered electric prostheses become more mainstream, more efficient, and lighter, many athletes may switch to a prosthesis that uses AI to reduce strain and improve balance. This could lead to these athletes having an incredible advantage over able-bodied athletes, specifically due to the further decreased amount of energy used by the body to support and control the limb. This will likely affect less competitive sports and result in bans on amputees from competitions such as marathons or high school and college sports. Future Consequences of Intelligent Prostheses“The human enhancement market will reveal the truth about our biological conditions — we are all disabled. ”

In the future, the features of all of these prostheses will likely be combined. An artificial hand that feels, sees, and predicts just like the real thing will be possible to implement within the space and weight capacity a prosthesis offers. Once the value of having these prostheses is close to the value of having a real limb, progress can become more focused on lowering the cost of these devices. But what happens when these devices become better than limbs while remaining relatively affordable? In the present, this is a controversial topic for many doctors. Some patients diagnosed with body dysmorphic disorder are seeking unnecessary amputations of limbs that they feel like are not theirs. This condition is typically met with little sympathy from surgeons, since it is a medical doctor’s duty to do no harm, and an unnecessary amputation will result in inconvenience at the least, with infection and regret being real risks of the surgery.

With advancements in AI technology in prostheses, it will become less of a regrettable choice and more of an opportunity for improvement. It is unknown how medical professionals will respond to this shift in demand for prostheses, but over time, it is likely that more and more people will be looking toward amputation as a solution to inconveniences like knee problems. It is also likely that the idea will gain traction in the body modification community for aesthetic reasons, as well as in certain industries where durability, strength, or precise motor control are valued in an employee. With the increasing capabilities of prostheses, it will become hard for people to ignore the appeal of them. Andy Miah, director of the Creative Futures Institute at the University of the West of Scotland, believes that in a few decades, people will be replacing healthy limbs with artificial ones (“The Future of Artificial Limbs”).