Building an artificially intelligent, open-source prosthetic leg

We’ve come a long way since the first prosthetic leg, and “smart” limbs, equipped with computing capabilities and artificial intelligence, are on the horizon. But for a team of engineers at the University of Michigan and Shirley Ryan AbilityLab, advances still aren’t happening fast enough. To move things along, they are giving away the plans to an AI prosthetic leg — hoping researchers will piggyback off each other’s work, taking prosthetics to new heights.

Mechanical engineer Elliott Rouse says he had the coolest job: traveling the world on a pro auto-racing team. But, after a few years, Rouse was no longer satisfied. He wanted to help people with disabilities. According to him, seeing people’s lives changed for the better is a “big deal.” So he shifted his focus to wearable robotics — namely “smart” bionic limbs.

The aim of such bionic limbs is to enable the user to go about their daily activities as an able-bodied person would. That is, more than a “dumb” prosthetic leg, smart limbs need to anticipate the user’s next move.

Imagine a robotic knee that changes the angle it bends when the user switches from walking up stairs to descending them. The difference is critical.

Now, instead of starting from scratch, researchers can use the Open-Source Leg as a common platform to build upon.

Regulating the motorized knee joint is a computer called the “controller.” Like an artificial brain, it figures out what is going on based on sensory information, then tells the prosthetic leg the position it needs to be in. There is a lot of pressure for engineers to perfect bionic limb controllers. One mistake (as in the case of the knee while stair-climbing) could lead to a tragic fall down a flight of stairs.

“The controller has to be right all the time. It gets minimal sensory input, and it has to decode what the person wants to do,” Rouse says, adding that designing the controller is a lot harder than creating the leg hardware — the structure, motor, and joints. The software has to be perfect, and the technology isn’t there yet.

“We don’t have one that has never made a mistake. That’s why you don’t see them in public,” Rouse says, adding that a few AI ankles are out in the “real world.” Knee joint controllers are harder to design due to the risks associated with maintaining balance and supporting the shifting body. A mistake in an AI knee has more significant consequences than an AI ankle.

Rouse says there are plenty of researchers working to advance controllers for bionics. The problem is that each group builds its own unique hardware — an investment in time and money. But it also means that it is harder for researchers to decipher the results of other teams’ published work and apply it to their own prosthetic leg designs. According to Rouse, the process is holding back the field.

“That’s time that can be used to solve real problems we have, instead of reinventing the wheel over and over,” he says. To speed things along, Rouse and his team built a plug-and-play bionic leg and made it freely available to the scientific community. Coming in at $28,500 — a far cry from the millions typically associated with building a bionic leg — they hope this will help researchers save costs and immediately begin testing different controllers on knee and ankle prosthetics.

Rouse, the lead designer from the University of Michigan, focused on keeping the leg simple and low cost, using inexpensive parts from a few suppliers. His flexible design can act as a knee, ankle, or both, with an onboard power supply. By using pancake-flat motors designed for drones, Rouse kept the knee efficient while boosting fine control for human-like movements.

The Open-Source Leg is intended as a common hardware platform that researchers can use to investigate controllers. However, Levi Hargrove, from the Shirley Ryan AbilityLab, developed an AI-based controller that he plans to soon make available on the website along with the leg — thereby also creating a starting point for control, and a baseline of comparison for controller research. The AI-based controller has an adaptable mode that can respond to the user’s actions, switching between activities like walking or stair-climbing. To do this, they used sensor data and signals from contracting leg muscles where the organic and bionic leg meet to predict the user’s next move.

Now, instead of starting from scratch, researchers can use the Open-Source Leg as a common platform to build upon. Open-source upper body limbs, such as hands, are already used by researchers, but this is the first such limb available for lower extremities.

Eric Rombokas, a University of Washington engineer, ordered the Open-Source Leg as soon as it was available. But he couldn’t wait for his order to come in, so he flew to Michigan to try the leg in Rouse’s lab, bringing his own controllers along with him.

It is difficult for small, new labs like Rombokas’ to land big funding opportunities that would allow them to build their own bionic limb. Plus, it isn’t his area of expertise anyway — his lab “isn’t a hardware lab” — so he says the Open-Source Leg is a “totally required thing to engage in the kind of research that I want to.”

“For years I’ve been interested in trying to take the methods that have been developed in machine learning and robotics and apply them in prosthetics, but I haven’t been able to show anything because there is no device to show it on. At the end of the day, you want to see the person with the amputation moving around in an agile and stable way and you can’t do that without a device,” Rombokas says.

As an open-source project, anyone can try out the leg and improve upon its functional design. Volunteer patients offer feedback, researchers publish their work, people share new insights on the Open-Source Leg website, and leg bionics improve. Already researchers at Georgia Tech EPIC lab and Northwestern University are putting it to use.

Rouse knows there are tradeoffs to making the leg open-sourced, especially losing the intellectual property and the ability to patent and commercialize a product. But Rouse says the leg isn’t a product. It is a research tool. If giving it away for free will lead to more exceptional outcomes for bionics as a whole, then Rouse is all-in. After all, his ultimate goal has always been to see people’s lives changed for the better, and that means sharing information in a way that facilitates impact.