Introduction
AIMO has high scientific standards and is closely associated with research. This is reflected not only in publications but also in the founding of AIMO. From the very beginning, we at AIMO wanted to create trust in the motion scan and score.
What is our claim?
We want to be thought leaders in our industry. To achieve this, we work closely with partners from the world of science. From AI to pain and ethics, we rely on excellent experts.
Discovery: What can the AIMO do?™ Motion scan?
Through the AIMOTM Movement scan and score detect momentary movement dysfunctions and muscular imbalances. Performing the overhead squat provides information about the following areas of the body:
- Shoulders and neck
- Spine
- Hands and arms
- Hip and thigh
- Knee and lower leg
- Foot and ankle joint
What does the score say about your movement?
Based on this, AIMO calculates the individual movement score. This is done by comparing it with the ideal movement. This is made possible by trained artificial intelligence. An ideal score would be 100%. Depending on the movement execution, most AIMO users initially receive a score of around 54%. One of the reasons for this is that this movement is unfamiliar to many and the individual movement sequences do not flow smoothly into one another. You often see evasive movements that deviate from the ideal movement. S
Recognizing and understanding deviation movements
- Movement of the knee inwards or outwards
- Lifting the shoulder
- Dropping the arm forward
- Squat cannot be performed deep enough
- Lifting the heels
- Lean forward strongly during the movement
- Moving the head forward
- Foot turns outwards
- Foot flattens
- Lower back curves inwards or outwards
- Body weight shifted strongly forward
- Body shifts to the left or right
In the app, you can take another look at your movements after each scan. This not only allows you to recognize your own weak points, but also improves your body awareness by mirroring your movement.
Technical evaluation of your movement
The score therefore evaluates the general functional movement quality. Your movement pattern is therefore made up of the combination and characteristics of your evasive movements. We explain possible causes so that you are not only shown your potential, but also understand what is going on in your body. It is important to us that our customers understand why an appropriate training program is being put together.
If you want to know more about it, why not try it out directly?.
Download AIMOTM download, scan your movement and start your individual training.
The beginnings of the AIMO™ motion scan
AIMO also started out small. However, our first motion scans were performed with fairly large setups. We presented the first scanning stations, consisting of an external 3D camera and a large screen, at trade fairs back in 2018. Before our dedicated AI team achieved the breakthrough in 2020 and AIMO went mobile, we trained our algorithms with 3D cameras and made great progress in digitizing the analysis of human movement. During this time, we have conducted a large number of experiments and scientific studies together with our artificial intelligence specialists. These have been published at international conferences and today show the steady progress of AIMO™ technology and the great successes of our team.
We would like to give you an insight into the development of AIMO and the various stages of development that AIMO has gone through so far. Here you can find out more about the AIMO technology.
Scientific projects of AIMO
1. interactive visualization of software metrics using multidimensional probability distribution.
- One of the authors of this article, which was published in 2018, is computer science professor Welf Löwe, who has worked closely with AIMO for many years. The article was published as part of the international VISSOFT conference (IEEE Working Conference on Software Visualization) in Madrid.
- He discusses the overall quality of software, which is made up of different quality aspects. This problem also had to be solved for AIMO™: the overall quality of a movement is made up of the quality of individual parts of the movement and the body parts involved.
- In general, quality can be broken down hierarchically into aspects that can be evaluated using various direct and indirect metrics. Conversely, quality aspects are summarized in order to evaluate the quality of an entire system.
- Different aspects of quality are recorded using different metrics and each of these metrics may have a different interpretation, scale, value range or measurement method. It is then difficult to automatically summarize (aggregate) these metrics. In order to investigate and compare different aggregation methods, the authors present a tool that can be used to visualize overall quality, quality aspects and metrics.
- The authors propose an approach to aggregate these metrics and quality aspects based on the distribution of (multidimensional) metrics. Through joint visualizations of these distributions, users can examine and compare the quality metrics of systems and their components, as well as identify patterns, correlations and anomalies. Furthermore, it is possible to identify common properties and errors, as the proposed visualization approach provides extensive interactions for visual queries of multidimensional data.
- The approach is applied and validated in two use cases. These are based on 30 real technical documentation projects with 20,000 XML documents and an open source project written in Java with 1,000 classes. The results show that the proposed approach enables an analyst to identify causes of poor or good software quality.
- However, the visualization approach is not limited to software quality and has already been successfully transferred to the visualization of motion quality and its aspects.
Find out more here:
2. introduction of quality models based on probabilities
- An approach for the mathematical definition of quality models was presented at the international conference for software engineering in Gothenburg in 2018. One of the authors is computer science professor Welf Löwe, who has worked closely with AIMO for many years.
- In this article, the authors present a mathematical definition of so-called quality models based on probabilities of multidimensional distributions of metrics. They illustrate their approach using a quality model with 30 standard metrics.
- Quality models combine several observed metrics into a common target metric. Each individual observed metric has its own interpretation, scale and value distribution. It is therefore not to be expected that a simple combination of observed metrics, e.g. by summation, will lead to an interpretable target metric and thus to a reliable basis for decision-making. The aggregation of observed metrics must be mathematically correctly defined and interpretable.
- On the one hand, the results show that metrics are generally not independent, as they measure different but correlated aspects of quality. However, they also show that a clearly defined and interpretable quality model can be designed.
Find out more here:
3. data-driven evaluation of human movements
- In June 2019, a paper by AIMO founder and sports scientist Danny Dressler, computer science professor Welf Löwe and senior developer Pavlo Liapota received the Best Paper Award at the International KES Conference. The scientific contribution appeared in the publication Smart Innovation, Systems and Technologies published by Springer Verlag.
- The article describes the automated assessment of musculoskeletal insufficiencies using a data-supported movement test.
- The quality analysis of human movements is already used in high-performance sports and in the diagnosis and therapy of musculoskeletal weaknesses. The authors describe five purely data-driven assessment methods for arbitrary human movements using low-cost 3D sensor technology. They evaluate their accuracy by comparing them to a standardized assessment method for deep squats based on human expert assessment. They explain the data-driven method, which shows high agreement with this standardized scoring method while requiring little expertise in human movement and no expertise in the scoring method itself. They also note that the digitized method enables effective and efficient, automatic and quantitative assessment of any human movement.
Find out more here:
4. on the way to an automated assessment of musculoskeletal limitations
- This article was also published in June 2019 in the publication Smart Innovation, Systems and Technologies by Springer Verlag. The article by AIMO founder and sports scientist Danny Dressler, computer science professor Welf Löwe and senior developer Pavlo Liapota presents a method for the quantitative evaluation of human movements.
- The highly complex algorithm developed for this method is the basis for the unique AIMO™ technology.
- The researchers used low-cost 3D sensor technology to record human movements. To measure the accuracy of the method, the assessments made by the method were compared with assessments made by human experts.
- The results show that the estimates of the presented method and those of the human experts are in high agreement. To achieve this, a new type of algorithm was developed that works for any time series.
Find out more here:
5. the effect of machine learning approaches in the assessment of human movements
- This article was presented at the 11th International e-Health Conference in Portugal in 2019. One of the authors is computer science professor Welf Löwe, with whom we have worked closely for many years.
- This collaboration also gave rise to the unique algorithms that are used today to calculate the AIMO™ motion score, among other things.
- The development of these algorithms is based on 3D sensor technology and machine learning approaches. These make it possible to create models for automated motion evaluation.
- In this article, the authors investigate the effect of different machine learning approaches on the accuracy of human movement assessment.
- As a result, it turns out that machine learning based on so-called „random forest“ models are best suited in this context. Since the results are not based on the movement specifications, they can help to improve the performance of automated human movement assessment in general.
Find out more here:
Hagelbäck J, Liapota P, Lincke A, Löwe W. The Performance of some Machine Learning Approaches in Human Movement Assessment. In: Macedo M, Rodrigues L (eds.). Proceedings of the 13th Multi Conference on Computer Science and Information systems (MCCSIS). Porto, Portugal. IADIS Press; 2019 p. 35-42. ISBN: 978-989-8533-89-0
6. variants of dynamic time adjustment (time warping) and their effect on the assessment of human movements
- This paper is the result of a collaboration between Linnaeus University and the Swedish software development company Softwerk and was presented at the 21st Conference on Artificial Intelligence in Las Vegas in 2019.
- The unique algorithms that today form the basis of the AIMO™ motion scan were developed under the scientific supervision of Linnaeus University and the professional work of Softwerk. The development began with the research and use of 3D sensor technology to evaluate human movements.
- This research work is based on machine learning approaches. These make it possible to create models for automated movement evaluation. The statements made by human experts about the quality of a movement are used to train the models. The algorithms learn from a large amount of evaluated movement data and will be able to evaluate the movement automatically in the future without the expert statements.
- The researchers and authors of this article investigated the effects of sequence alignment on the accuracy and reaction time of motion assessment. The influence of variants of dynamic time warping (DTW) on the accuracy of the evaluation assessment is analyzed.
- The results show that an automated truncation of frames not belonging to the motion (using a DTW variant) followed by an alignment of the selected frames of the motions (based on another DTW variant) outperforms the original DTW approaches.
- Since these results are independent of the selected learning approach, they can also help to improve the performance of automated motion assessments in general.
Find out more here:
7. compatibility of different 3D camera technologies
- An article on the use of 3D sensor technology to assess human movement was published in 2019 as part of the World Congress on Informatics, Computer Science and Applied Computing. One of the authors is computer scientist and AIMO co-founder Eduard Rall.
- An important basis of AIMO™ technology is machine learning. Machine learning is used to develop models for the automated evaluation of movements based on a wide range of different movement data. If a new camera technology does not match the technology with which a model was learned, this model cannot be safely transferred to this new camera technology. A new model for automated motion analysis must then be created for this camera technology.
- The authors therefore present a cost-effective method that can be used to check whether a particular 3D camera technology is compatible with another. They apply the method to the Kinect, Astra Mini and Real Sense camera technologies as examples and find that these technologies do not match. This means that the motion analysis models developed for the Kinect camera, for example, cannot be used for the other two technologies without losing precision.
- The method presented here can save time and effort when new camera technologies are to be integrated into an existing evaluation environment. The method can also be used in areas other than motion analysis, as it is valid independently of movements and cameras.
Find out more here:
You want to test all this in reality yourself? Then get started.
