Smart wheelchair

Assignment for the Course Data Driven Design


Team: Carolina Falcão Duarte, Salvadore Luca Cucinella,          

Sanket Mane, Paola Montserrat Bautista and Koen van der Loop


My main activities: Context research, renders and animations, data flow design, adaptation loop, reflection on ethics.

Data and design 

How to design an adaptive wheelchair?

An exploration on using data and machine learning as design materials

This work was developed during the Designing Data-Driven Products and Services for the Internet of Things course at TUDelft in 2018/9.

The challenge of this project  was to come up with an IoT Product-Service-System that is adaptive and can evolve through time using machine learning. In addition, we were asked to take a brief look at the context without getting into a deep research regarding the user group.

Our design took into account children who have paraplegia or other disability that harms only the legs movements. .


The context

According to Li et al. (2017), regular physical activity in children with physical disabilities is important for their current and future health and well-being. The researchers argue that children with physical disabilities are mostly sedentary, and, to avoid the risk of decreased physical functioning and additional health problem, they should have at least 60 min of physical activities daily. 

The  target group: We analyzed and decided to tackle the condition of children between 8 and 12 years old, with a paralysis that affects parts of the trunk and legs, paraplegia.

The Goal:  To help children with paraplegia by improving their physical condition. A good physical condition could make these users able to deal with everyday difficulties (i.e. overcome high inclination ramps). 

an evergreen product

The concept and its adaptation loop

Wheely is a wheelchair designed for children with paraplegia, it measures and monitors the physical improvements of the user when he/she interacts with the UI.

A machine learning (ML) system is supervised by a physiotherapist and provides recommendations about new possible games/exercises in an attempt to improve (i.e.) their muscle strength without depending exclusively on a visit to a physiotherapist’s office.

The wheelchair becomes a partner, by providing playful exercises. For the physiotherapist, it provides a way to assist the patients outside the office.

future scenarios

The evolution of the Product-Service-System

The system offers exercise programs that are reviewed and improved  by the physiotherapist.

The exercise suggestion becomes more accurate as more users engage in the system. New thematic exercises are introduced.

The physiotherapist’s workload becomes lower, although this professional involvement remains the same when it comes to deciding which exercise is better for the patient.

Users can interact with each other through the games. They can choose to play in the same place or at their homes. Connecting children through games/exercises can impact the psychological state of those users. A data-driven wheelchair might promote “the social influences in motivating active behavior in children” (R. Li et al., 2017).

technical aspects

The system’s architecture

The Algorithms

The algorithm measures the level activity of the user (that is the level of physical function of a person) and gives a score (points) between 0-100.
The points are influenced by the movement, heart rate factor, adaptation factor and the age factor (a) of the user.

Heart Rate Zone (HRZ): It is the ratio of heart rate of present time to that of maximum heart rate. Very light (50-60%) light (60-70%), moderate( 70-80%), hard (80-90%0, maximum (90-100%)

Adaptation factor: It depends on loss in muscle strength over some period of time. It is the ratio of time required to lose some specific amount of muscle strength to that of time taken by the average user to lose the same strength. For example, the system can compare the current score with the user score from a month ago.

Pressure Sensor Data:  It gives information of the duration the user is doing the exercises. Each exercise if done properly will have a unique pressure variation pattern. The machine learning algorithm detects this pattern and infers from that data the duration the user is doing the exercise.

Age Factor (a) : The activity level would be influenced by the age of the children. The older children would perform few exercises easily than younger children. Hence it is necessary to include age factor in the algorithm.

Effective Time (t1)= It is the time (in hours) which includes pressure sensor data, heart rate data and muscle strength data to calculate the effective time the user is doing the exercise.

f(Pressure sensor data*Adaptation factor*Heart Rate Zone)

Activity Level = (a*t1/24)*100

The Recommendation System

The first step to recommend new exercise to the user would be to identify the user itself. It can be done with the 4 factors i.e Recommendation (R), Exercise(E), Activity Score(X), Time(T).
Exercise (E) -There are many exercises which the user can perform but to understand the content let’s consider four exercises Exercise 1 (A), Exercise 2 (B), Exercise 3 (S), and Exercise 4 (F). Then there would be different possibilities as shown in the table. A number corresponding to that exercise will show to indicate the exercise the user is doing

Recommendation (R): is the recommendation given by the doctor which is dependent on the type of injury the user has. Every user will be assigned a corresponding number for it.
Activity Level (X): is the points the user receives on the level of activity scale.
Time (T) is the total time while he is doing the exercises, therefore a user can be represented by U = [ R, E, X, T)
The machine learning algorithm uses collaborative filtering to suggest:

1. A new exercise for users with the same recommendation  and activity level.
2. How much time a user has to spend to improve its lifestyle for users doing having the same exercise, recommendation and score


Sharing health-related data is a delicate matter for those responsible for the infant and for the child. For that reason, we decided to provide a different way to communicate those issues in a way that can be understood by the child as well.
The consent form is just a rough idea that would need to be evaluated, tested and iterated with potential buyers and users to be as clear and as informative as possible.
We used the Contextual integrity (Apthorpe et al., 2018) as the main reference for the proposed privacy norms policy.

User control strategies

Before Buying

The information about data collection, storing and sharing is presented as one of the features of the product-service system. It gives the user the opportunity of taking into account what is done with the data and ask questions before purchasing the product.

Purchase moment

The parents are asked if they consent with sharing this data with the company after the first use of the product.

First use

The parents receive an email informing them about the granular data consent policy.

Every 6 months

The parents receive an email reminding them about the data consent. The e-mail states, again, that the data will be stored for 5 years, who will look at it and how can they have control over this data.

Feedback Every Year

The company sends an email and informs in the website what improvements were done thanks to the users who shared their data. The company informs clearly those improvements could benefit the kids who use Wheely. Members of the company are shown in videos explaining how they use the data and how it was important to improve their work and deliver value to the kids.

All the time

A feature in the user profile allows parents to see who has accessed the data and when.

mock-ups Exhibition

As a final exercise, we developed a mock-up of our design. We showed where to position the sensors and exactly what kind of sensors we aimed for during our system design.


A brief reflection about this work

This project allowed me to use a different mindset. The sensors and the data that we could take out of a wheelchair were the starting point, while I am used to taking the user as the starting point.

This design has improved my understanding of the Internet of things and the responsibility of designers to build the bridge between the user and the system while observing the ethical issues involved.