Symbionic Postmortem
Reflections and admittance of failure.
It's been almost 2 years since my last post in the Symbionic project. If you found this project just now, let me briefly explain how we failed. You may benefit from it.
The reason it took me 2 years to write this post is the same reason I stopped the project work: my wife suffered from a burnout. It affected our entire family and I decided to step down from all my personal projects. I also took extra leave from work, all to support my wife in her recovery. She has been slowly, but surely, recovering and now I find myself with some extra time and energy again. I wish to somehow continue or restart this project, but first I must look back and reflect.
From the original goals it's clear that we failed them. We already weakly admitted to that in another post. We did not succeed in creating an affordable and easy to use gesture detection system for bionics or other purposes. We did learn a lot and through the process I am definitely a better, more skilled person. If you wish to learn a lot, then this topic is an amazing intersection of technology and societal impact. Let me recap some of our hurdles, so you may try to overcome them.
The biggest technical issue we encountered, next to finding good bio-sensors, were the software connection aspects. There are so many sensors and microcontroller boards out there, all with different and hard to use interfaces. We struggled with our limited embedded software knowledge to integrate whatever sensor we wanted to use. Fortunately these last two years brainflow.org was started and has been tackling this exact problem. I've kept an eye on it over the years. You should also check it out right now, because based on my experience I say we are looking at the future of realtime bio-sensor data analysis.
Another tech issue was writing software that is both fast enough for the high sensor data rates, as well as easy to write and extend for our desired machine learning purposes. This is a very common issue called the two language problem. You'll mostly write in a slow running, but high level, language like Python to move fast in your project and then slowly recode into a fast running language like c/c++ for performance. This blocks me every time. I had no hopes for solving this issue, until I found the Julia language. Julia is growing in popularity precisely for solving the two language problem, is already used by most computational scientists and is rapidly expanding into business and industry. I've quickly taught myself Julia programming.
The other struggle was finding collaborators and integrating their knowledge. There is a lot of academic work going on in neurotechnology and bio signal analysis, but a lot less applied projects. There are quite a few amateur enthusiasts, but they struggle to collaborate through space and time, many already failing before another starts. I've also seen a dozen startups come and go now in the last 4 years, all their efforts and knowledge lost in time. We did encounter a few absolutely amazing projects trying to build a bigger platform or community. One of those may be the seed of future success.
For example, NeuroTechX is aiming at building a community of neurotechnology enthusiasts. I have not been actively involved and it still seems like a lot of loosely coupled people. However I will continue to monitor their progress.
For making open source bio-sensors controllers, openbci.com is still courageously struggling out there. Their boards are multipurpose for use with EEG, ECG and EMG sensors. I think this leads to their boards still being a bit in the expensive range for personal projects, but there is nothing else with their vision out there. I'm currently looking into buying their starter kit.
OpenBCI also struggled with the software aspects of the work and has therefore teamed up with BrainFlow. This may turn into a potent biosensor platform, where open source technology can lead to compounding effects. The technical feasibility of BCI products has already been proven by many dead startup projects. The problem is creating a sustainable organization around it. OpenBCI and BrainFlow may be the seed for that organization. And if not, then all their knowledge is at least open sourced and can readily be used by other startups.
A more general issue you will encounter in any project is the human factor. You will have to deal with humans and work with them in order to succeed, with all their good and bad parts. I struggled tremendously to keep myself and others motivated. Very often, all you need to succeed is enough persistence. But how to gain that? To combat this issue I have been learning a lot about human psychology, through reading books, self reflection, meditation and by following the rationalist community. For engineers it's all too easy to ignore the human aspects and focus on the technology. Please don't make that mistake! Focus on improving your own mind and find and improve those great minds around you.
The field of neurotechnology is littered with the corpses of failed startup projects. A lot of precious knowledge is lost forever. I am happy I documented our project online for others and hope it helps you. But we need a better way for others to build on top of all those failures. We need a beacon to gather the enthusiasts and compound our knowledge. For the open source bio-sensor stack, I believe OpenBCI+BrainFlow is that beacon. I've already joined BrainFlow as a contributor and if you find this post I hope you too will check it out too. Feel free to contact me if you have other ideas, I'm most active on LinkedIn. Signing off from Symbionic.
