Throughout the last semester I participated with a couple of fellow students from CDTM at the Tech Challenge of unternehmerTUM / Technical University Munich. The exciting thing about this course is that it is in comparison to a lot of lectures or seminars very hands-on. The goal of the Tech Challenge was to create, prototype and test a solution jointly together with a corporate partner. The whole project had an entrepreneurial touch tailoring the learning experience to a possible entrepreneurial journey with a final pitch of selected teams at the demo day. It was especially a lot of fun pitching our project and prototype on stage with the team.
Our team specificly tackled the AI & ML Challenge, one of four possible tracks that teams can choose. The task was to somehow leverage machine learning technology to build a product in the sphere of the project partner that could be relevant. Our project partner was neurotrim, a healthcare startup providing a novel method of analyzing and improving your body by using a combination of software and hardware. Their hardware device basically enables guided training and tracking which can be assisted by a therapist or trainer.
The Idea: feasio - Virtual coaches. Reinvented.
What we came up with is feasio. The basic idea was to provide a software application that can act as a virtual therapist or trainer, which can guide on the one hand therapeutic training sessions and provide on the other hand real-time feedback on the execution of the exercises. Such a product would address different problems. Generally, in physiotherapy sessions, physiotherapists explain to their patients exercises that they should perform at home, as just attending the sessions is not enough. To obtain long-term results, it is crucial that these patients perform these exercises at home, and in the right way. However, many physiotherapists do not provide their patients with an easy overview of exercises or an exercise schedule. Thus, patients exercising at home tend to forget the number of repetitions of specific training tasks or how to perform the tasks. Furthermore, while exercising at home, patients do not receive any kind of feedback. Patients do not know if they performed the exercise correctly, how they should adjust their movements, or how their performance changed over time. Similarly, patients often lack the intrinsic motivation to properly abide by the instructions and schedule set out by the physiotherapist, and is dire need of some form of gamification to make their exercises interesting. People training with a personal trainer can also relate to these problems.
Feasio wants to tackle these customer needs with a web app that provides an exercise schedule, exercise reminders as well as clear exercise instructions through a gamified approach. Using a webcam or smartphone camera, every exercise could be tracked by an algorithm and real-time feedback could be provided. Patients would know if they are executing an exercise correctly at any time and receive “feasio health points” for completed exercises to keep their motivation high. This product in the realm of the physio and fitness market can be targeted at physiotherapy patients and fitness enthusiasts to complement their existing training.
Prototype: Leveraging Computer Vision / Pose Estimation
For the sake of our prototype, we implemented a web application with simple HTML, CSS and JS. As time was very scarce and the goal was not an production ready system, we didn't want to overcomplicate our technology stack. As this was the AI/ML challenge, it was further expected to involve a form of this at the moment present and hype technology. In particular, we used a computer vision method called pose estimation. Pose estimation aims at detecting different body parts of a human based on images or videos. One breakthrough in real-time multi-person pose estimation was achieved by researchers of the Carnegie Mellon University published in 2017, where they managed map body parts such as joints pretty accurately in real-time to a video using two convolutional neural networks. This accelerated the research and progress in pose estimation using computer vision methods.
By means of pose estimation, we wanted to be able to track a persons execution of an exercise, guide the exercise and give feedback. In a fully developed scenario, the algorithm would detect the body movements coming through the webcam, match it to an ideal movement and provide the subject with improvement suggestions regarding e.g. posture. Further, the application would provide the subject with a training plan and video tutorials to follow the exercise. Within the scope of the tech challenge we weren't able to implement all features. But in particular the prototype aimed to validate the business idea.
A link to our git repo can be found here: https://github.com/patricktu2/Feasio
Demo Day: Pitching the idea
Even though it is called the Tech Challenge, there was also a lot of business stuff involved in the learning process. Part of the deliverables was besides a prototype, formalizing the idea into a business plan, designing visuals and presenting the idea. At the demo day our team was one of the 8 teams chosen to present our idea on stage in front of the audience.
Even though we didn't won the challenge, it was still a very good experience participating. Having a very interdisciplinary team, I was still the most techie-like team member. I wouldn't consider myself as a whole techie, but this was definitely a good exercise on my journey to become one. While my team covered the business stuff such as validating the business idea, conducting user interviews or writing the business plan, I tried coming up with our prototype. While I certainly love trying out new technology, I still feel I just scratched the surface of the technology we used. I am already very excited for my next CDTM course Managing Product Developments, which is a course also focusing about developing a product idea including a prototype. There I'll have another chance to apply some of my learnings and get deeper into the tech!
Thanks to the wonderful team Centerlings: Tomas Li, Leon Szeli, Özgur Güzel (and me). All work presented here was jointly developed.