I recently reconnected with Peter Kaempf, a member of the innovative creator community on the HWTrek platform, who I had met at SXSW in Austin last year and who my co-workers ran into at the Intel Partner Summit at Embedded World in Nuremberg. Peter is Head of Special Product Development at STABILO International GmbH. Fortunately for us, he was keen to talk with us about the development of Digipen and so provided a wonderful interview. Enjoy….
Peter Kaempf and HWTrek CEO Lucas Wang (Intel Partner Summit, Embedded World 2017)
Please introduce yourself and your project?
The Digipen is a sensor-enhanced writing instrument with internal data processing capabilities and an external data link for communication with compatible devices. It will register accelerations and its position in space and correct this position data for drift. At the same time, it can be used as a regular ballpoint pen, on regular paper. Motion data can be stored in 64 Mbit of internal memory or transmitted via a BLE connection to a connected device. Myself, I directed the STABILO pen development since 2000 and have a background in aerospace engineering. We have been looking for ways to connect handwriting and computers for years, and a few years ago I decided to jump into this venture full time. Thankfully, the company trusted me to do this, so I assembled a small team of engineers and started the development of the Digipen. Including myself, we are now 5 people working on hard- and software. On the sales side, we currently use an external consultant and are in the process of building up our own expertise.
When we first met, you were demoing Digipen at SXSW 2016 in the German Haus – very much in the spirit of a startup, something a bit unusual for a large, more than 160-year-old company. Was this “startup approach” a calculated part of the product development process and marketing strategy for Digipen?
Yes, absolutely for the development. While the company culture is a valuable part of STABILO, it is less well suited to a geeky, risk-taking undertaking. I figured we need a different culture to become successful quickly, at the price of an increased risk of screw-ups. Yes, we had screw-ups, but they were limited, so I am very happy with the outcome. On the marketing side, I hope to profit from the high brand awareness STABILO enjoys in Europe, and the SXSW demo was rather unusual for our general approach.
What’s the inspiration for your project? What problem does it solve or address?
We study the importance of handwriting and see that kids today have less handwriting proficiency than the generations before. You can’t blame them – with all the digital distractions, there is simply not much time left for developing the fine motor skills required for good handwriting. In school, this becomes problematic: The curriculum expects them to have a high degree of automation in their handwriting from fourth grade on, so they can take notes and follow the topics in parallel. However, when this automation was never learned properly, the process of handwriting will absorb too much of the kids’ cognitive capacity – they fall behind and cannot follow the teacher anymore. With the Digipen, we hope to develop a measurement tool so teachers and parents can see how far a kid is on the way to fully automated handwriting. We also develop exercise books to train them, so we work to get a full solution in place that will address the handwriting crisis in schools. In Germany, we see severe handwriting problems in more than half the boys already, so this is a very serious problem that has slowly grown bigger and bigger over the last years.
What are the applications & use cases for Digipen? Do you have any interesting user stories to share around these applications?
The primary application is the measurement of the degree of automation in handwriting. But there is more:
Plot examples of saved sensor data from the Digipen
Since the pen can measure acceleration and angular velocities around all three axes and updates its attitude information every 5 ms, it can be used for motion tracking in 3D. The internal force sensor can be activated by an optional button, so a proportional, user-selectable scalar parameter can be added. This allows the pen to be used as an ergonomic computer mouse with force-sensitive input or as a 3D motion controller in a virtual or augmented reality system.
For users with developed handwriting skills, the sensor signals of the product can be fed into a pattern matching algorithm that will run on the connected device. The sensor fusion in the product will normalize the sensor readings, subtract drift and reduce the data volume such that the data link to the connected device does not need more energy than necessary. On the connected device, individual characters are modeled by Hidden Markov Models (HMMs) and concatenated to word models. A statistical language model is used to enhance recognition performance and restrict the search space. The resulting text output can be displayed and stored on the connected device. This is the version I demoed on SXSW.
One of the more interesting use cases, aside from education, for Digipen, is medical. An example is using AI to compensate for the hand movements of users with Parkinson’s. Can you speak more to this and other medical use cases? I’ve read a fascinating report from MIT about using digital pens and AI to help detect early signs of dementia – Alzheimer’s, Parkinson’s, etc.
Yes, the Digipen can be used to diagnose Parkinson’s or even schizophrenia. Our first customer uses it for Parkinson diagnostics, and it can be used to fine-tune the cocktail of medication each Parkinson sufferer needs to reduce the tremor as much as possible. For tremor compensation, we would need some active movement by the pen, which we have not implemented. For us the medical applications look a bit scary, with all the regulation and certification hurdles, so we want to focus on the school market right now. However, we are happy to produce the pen for a partner who wants to sell it as a medical device.
What solutions did you use for hardware design?
First, I took apart all my computers, laptops, digital cameras, and cell phones to see how others do it. Then I made a very crude first design that used a circuit board wedged between two halves of a split wooden rod, just to see how the signals looked. The next step used a cylindrical 3D-printed body and then when the internals could be defined better, I got help from a German industrial designer to finalize the outer shape and the inner layout.
What solutions did you use for prototyping?
The first circuit boards were prepared with the help of local companies for small-scale circuit board production. I wish I had learned of HWTrek’s website earlier – when we met in Austin, most of the prototyping work had already been done. But for the initial small production volume, I still needed someone who could economically produce the injection-molded parts of the pen. STABILO does a lot of injection molding, but on a much bigger scale with multi-cavity tools, which would have been much too expensive. So I was happy that I found two partners in Asia for some of the injection molded parts that we now use for the first production use of the Digipen.
What resources have you used for sourcing and supply chain management?
Being located close to Nuremberg made the embedded world fair here the first destination. This was a great start into the field, and the next step was a cooperation with several local companies, motivated by public funding. With the combined experience we could cover the first years well. But I remember that it took one year until I found a force sensor small and sensitive enough for my demands.
For the production of circuit boards, we have started a cooperation with a Japanese producer of electronic components, so this part of the supply chain has been outsourced. The non-electronic parts are sourced locally.
What were the most difficult things to source for your project and how did you source them?
The force sensor is unique and was not easy to find, and also the battery was initially hard to source. However, two years later the selected model is offered by a wide variety of companies – I was simply a bit ahead of the curve.
What tools, if any, do you use for real-time collaboration on your project (with team members and partners)?
Email is far and out the most important. I also use video conferencing and the telephone, but with my busy schedule, the asynchronous nature of email is ideal.
What have been the significant challenges or obstacles you’ve faced on the project? How were they resolved?
The biggest challenge was and still is my ignorance. I am still learning something every day. Now, this may sound like a platitude, but there have been many face-palm moments where in hindsight all was so obvious.
A big challenge is also the time it takes to develop good software and the time to produce good injection molding tools. Fortunately, the contacts I got through HWTrek could produce good parts in a fraction of the time the molds would have taken here in Germany. Thankfully, my experience with injection molding helped to make the contact with the Asian partners very smooth and pleasant.
What are the takeaways and lessons learned from working on this project that you’d like to share with other IoT hardware developers?
Stay with standardized solutions and simplify things as much as possible.
What advice would you give yourself if you could go back in time to the beginning of your product development?
Oh man, this would be a long list! I would not stop talking to my earlier self for days. It is impossible to compress all the interlocking experience in a memorable phrase. However, one short advice would be sure: Look at HWTrek and get some help!
If you are planning a new version of your project or future devices, what will you do differently?
The publicly funded research program was not terribly efficient but at that stage the right thing to do. A future project will probably do without it. A new version is already in the planning, and it will use fewer components. However, this is only possible with new parts becoming available now. For the future samples, I intend to turn to Asian sources immediately – with my HWTrek contacts I see this as the best way forward.
We all know the phrase “hardware is hard.” Is there something that was much easier than you had initially thought when you started out on your hardware journey?
Yes, hardware is hard, but it can be measured and observed. Software is sometimes much harder, especially when an unanticipated effect shows up only rarely and by chance. What went really easy was the contact and communication with far-away partners whom I knew only per mail.
What trend do you see that is changing your sector/industry or what shift would you like to see happen?
The trend is clearly modularization and increasing complexity. Everything from communication protocols to legal regulations has already reached mind-boggling proportions, and there is no sign of stopping. We should not let lawyers and accountants make the rules, but makers and builders. In other words: Let pragmatic people with a vision decide, not bureaucrats.
What’s next for your project?
We need to lower the BOM and widen the applications. In order to help third parties to develop for the Digipen, we are going to add libraries for letter recognition.
And now for something completely different, fun questions….
What are your ‘go-to’ sources for tech information and news? (Do you have any recommendations for a must-read/watch/listen to article, book, blog, film, or podcast, etc.?)
Whoa – that is harder to answer than it sounds. When it comes to tech info, I absorb news like a sponge and have a hard time later to tell where it came from. Obviously, sites like StackExchange and Slashdot should be mentioned here, but also personal contacts. I am fortunate to know an electrical engineer who is simply a genius – he helped me with the first samples of the Digipen, and I still talk to him regularly when I have new questions. Google only helps when you know what to ask for, but with a friendly expert who knows your situation, you will find a solution in minutes where without him it could have absorbed weeks of research.
What’s currently on your playlist, what are you listening to these days?
Classical music. Everything from Bach to Brahms. My playlist also includes Tchaikovsky or Vivaldi, and the most modern composer on it is Philip Glass.
Philip Glass (photo credit: PhilipGlass.com)
What fuels you (coffee, tea, or….)? When you’re low on creative juice, what is your #1 method to get back on track?
I consider myself an introvert, but I get motivated most by talking with others. It helps to verbally express a problem I have, and explaining it to someone else will also help me to understand it better. The reply will lead my brain on a new track, and when the conversation takes its course, it will result in new viewpoints and new ideas very quickly.
What do you recommend (place to go/see, what to eat) for a visitor to Palo Alto?
I was once in Palo Alto to fly from the local airport there. Was a fun time, and I still have a deposit at the flying club. But I have an odd taste of what to see – most memorable for me were places like residential areas of Tokyo, a nightly walk through Shanghai or a slum in Kenya. The biggest impressions I got from unanticipated and novel situations. So go where tourists will not!
Sorry, Peter, I’ve been recently prepared too many interview questions for a bunch of members of our community – both creators and experts – and as a result, they are intermingled in my thoughts. I, of course, meant to ask you about Erlangen and Nuremberg.
Considering Erlangen and Nuremberg – you know how it is: You never visit the spectacular things where you live because you always can do so tomorrow. Only when a friend from somewhere else visits you, you show him/her your town and get a chance to see it with other eyes. So I consider myself to be a poor guide of what to see around here.
By H. Helmlechner (Own work) [CC BY-SA 3.0 (http://creativecommons.org/licenses/by-sa/3.0)], via Wikimedia Commons
A time machine, of course. Next would be the Star Trek computer that really understands voice commands.
