26 May

Identifying new ways of describing behavioural cues for SAM

In SAM, our objective is to identify attachment patterns to eventually categorise the attachment status of a child given a caregiver. By doing so, we will be able to support medical practitioners to focus their efforts on children that needs attention.

In order to identify attachment patterns, we are looking into behavioural cues from children when they play with our SAM game. We think that one of these cues could be hidden in the way children are manipulating the dolls while they enact their story. To verify this hypothesis, we had to look for a non intrusive mean of observing how children manipulate the dolls in the game.

There are several approaches we could follow. For example, we could measure the position of the dolls into space by analyzing the video recordings we are already collecting for facial behavioural cues. However, this option has many drawbacks. First, it is difficult to get continuous location of the dolls as they may be hidden by children’s hands or the furnitures on the mat. Another challenge is that this approach would require specific computer vision algorithms to identify the dolls in the video recordings, which would take time to design with unpredictable results at the current stage of the project. A more promising option would be to collect data directly from the dolls and get around the limitations of the computer vision alternative.

A set of familly wooden dolls

A set of familly dolls

While sounding great, collecting information directly from the dolls rises many questions. To begin with, what the dolls should tell us? There are so many different type of data we could collect and analyse. For instance, like in computer vision, we could measure the spatial configuration of the doll in space. Measuring the pressure applied on the doll during the game could be another behavioural cue. We could also collect bioemetric data such as skin conductance or heart beat like many smart watches and sport bands already do. Unfortunately, time is a critical resource and in this project we won’t be able to investigate all the possible ways offered to us.

Then, the next question would look into what sensors to integrate to our toys. Fortunately, the high constraints of size and autonomy has guided us to get a good idea of what we needed. Finally, how the dolls should talk to us? This last question is as essential as the previous ones. The way the data are transfered from the dolls to the computer could affect the game experience. For example, if the dolls were connected to the computer using a cable, children would not be able to move the dolls freely during the game degrading the game experience overall. However getting rid of cables is not trivial. This raises again other questions: should we transfer data to the computer as the children are playing the game or should we store the data locally in the dolls and then transfer them to our secure storage server after the children are done playing? Cableless also means that our toys would need to be selfpowered. But what would be the impact of either data transfer solution on the energy consomption and the battery integrated to the dolls?

This post gave a quick overview of the challenges to take into consideration for designing artefacts to help us investigating new ways of measuring attachment. The next post will bring a few insights to overcome these challenges with a preview of our smart dolls prototype.

25 May

Data collection of the first phase completed!

We have finally collected over 60 samples! While this number does not seem to be very large, it required a lot of efforts to reach this milestone. Because we are designing an automated attachment measurement tool, we need to measure attachment with the existing psychiatric assessments to compare the results with the results of our SAM tool. Therefore, each child in our study had to play with us twice with 8 weeks of intervals between either MCAST or SAM assessments.

Another challenge was getting into Glasgow’s primary schools to collect data. The study requires a lot of equipment such as a large black screen, some tripods, some video cameras, a big laptop, a few computer accessories, some toys and a doll house. Bringing all these items with us each time is laborious as we collected data in several schools at the same time.

It is worth noting that we could not have reached this milestone without the support of the parents and the schools we are visiting. We would like to warmly thank the parents, the children and all the kind teachers and staff members at schools that helped us making the study organisation straightforward.

SAM setup: SAM setup: the children tell stories to the computer using the two smart dolls

While it has been an intensive effort, we were delighted by the children’s reactions to our study. They were all so excited to participate and they left our experiment with joy and a big smile everytime! There were some children asking in the school when they could try our games and even some that had done the study came back later in the day to tell us how much fun they had playing with us! These little things also help to keep us motivated to knock out as much work as we can to complete all the tasks that are left.

So, where do we go from here? In an immediate future, we will start analyzing the data we have just collected, manually, as psychiatrics do, and automatically using computers and artificial intelligence. In order to make the computer smart enough to measure attachment, we are going to design some specific algorithms that will use the data we have collected to determine the attachment status of the participant. This will greatly help doctors tofocus on patient in need as we hope that SAM will identify the different attachment categories. We have also collected feedback from children about our game. All the feedback were great and it will allow us to make SAM even better for the next round of data collection!