Opinions & Editorials Standards & Regulations

AI Tackles Exoskeleton Categorization in 2023

Exoskeleton Taxonomy 2023

Last week the 11th ASTM International Committee F48 on Exoskeletons and Exosuits meeting was held in conjunction with ASTM committee week this April in Denver, Colorado. Coincidently, this also marked 125 years of ASTM International. Dozens of committee members discussed a plethora of exo-related subjects, including a surprise revisit of exoskeleton categorization, classification, and taxonomy.

Why Categorization Matters?

Organized shelf at Whole Foods Market. Lets do the same for the exoskeleton industry!

There are two ways to look at sorting and partitioning wearable devices. One is from the user side, and one is from the producer side. From the consumer point of view, it is just confusing to talk about all exoskeletons, exosuits, dermoskeletons, wearable robots, etc… all at the same time. This is the equivalent of trying to buy a transportation device without an idea if it is a car, truck, bicycle, plane, or boat. It is just easier to break down the wide range of devices into groups and subgroups. The last time ExR visited this topic was in 2016: Making Exoskeleton Information More Accessible. Our example at the time was the Whole Foods juice bar, which back then, was pleasingly sorted by volume and ingredients using color and pictograms (right).

A second way to look at exoskeleton taxonomy is from the point of view of producers and developers. Without a clear categorization, it could become confusing which standard (or future certification) applies to which device. Committee F48 already has 22 published standards, and there are more published and being written by AFNOR (French Standards Association Committee X35A on Ergonomics: Occupational Health and Safety), DIN (German Institute of Standardization Committee NA 023-00-08 GA on Exoskeletons), CEN (European Committee of Standardization, Working Group WS EXOSK Integration), ISO, and IEEE. If having so many standards organizations is not enough, exoskeleton classification is also being challenged by the large number of new devices that are being introduced to the market.

Does Chat GPT Have Anything to Say About It?

One idea that was discussed at the F48 meeting was to run an AI language model through the already existing exoskeleton device catalog on the Exoskeleton Report. The catalog was exported to .cvs table format, turned to long form text, and uploaded to ChatGPT 4.0. Unfortunately, only the short description for each device was uploaded due to unforeseen input length limitations. Both current and retired wearables were used in the analysis. After some prompting, this is the exoskeleton taxonomy that was created:

  • Mobility Boosters
    • Paraplegic Assistance
    • Gait Assistance
    • Stair Climbing
    • Knee Support
    • Hip Support
  • Rehabilitation Innovators
    • Lower Body Recovery & Gait Training
    • Upper Body Recovery
  • Occupational Enhancers
    • Industrial Support
      • Lumbar
      • Hip
    • Military and Law Enforcement
    • Energy Harvesting
    • Skiing
  • Pediatric Solutions
    • Gait Rehabilitation
    • General Rehabilitation
    • Mobility Assistance
  • Hand Function Helpers
    • Grasp Assistance
    • Inflatable Support
    • Hand Rehabilitation
  • Customizable & Modular Exoskeletons
    • Modular Rehabilitation
    • Single Joint Type

“This revised taxonomy keeps the focus on primary use cases, target demographics, and unique features while simplifying the categorization of devices,” adds ChatGPT 4.0 at the end of its output.


Two interesting choices in the output above are probably good enough to implement right away. First, calling “industrial” exoskeletons “occupational” is likely a better term. Committee F48 is already moving in that direction, and interestingly, the AI model chose to make the switch without being presented with that information. Second, separating pediatric exoskeletons into their own category is probably a good idea too. Right now, pediatric devices are labeled with a tag or modifier on the Exoskeleton Report, and treating them as separate entries may make it easier to highlight them.

Otherwise, the classification model provided by the AI is not bad for a first attempt but is a bit confusing. Unfortunately, the exoskeleton classification created in 2016 for ExR is also showing its age (see: Introduction to the Commercial Exoskeletons Catalog). Over the years, a top-down filter approach seems to have worked well, starting with all devices and filtering down based on application, energy source, and the human joints targeted by the wearable, but that is not the same as a taxonomy or a classification model.

In conclusion, the world of exoskeletons has evolved multiple times through the years. At one point, “biomechatronic” devices was the dominant classification. Then it was switched out by “powered exoskeletons” until we saw the rise of passive (elastic) ones. There is no telling what terms will be used by users, creators, or evaluators in the future, but hopefully, it will make it easier to have a productive conversation about exo technology.

Add Comment

Click here to post a comment

Upcoming Events