So it begins

With fund secured, we are very happy to start the work on this research project. And we want to say thank you for everyone who has made it possible, especially ESP's Science Leads: Maddie Cusimano and Sara Keen for having trusted us with one of the AI for Interspecies Communication grants.

Our project scope begins with the throughout understanding of the "Movement Sensor Dataset for Dog Behavior Classification" (Vehkaoja, et. al., 2021), which has been already very insightful.

In the "Description of movement sensor dataset for dog behavior classification" (Vehkaoja et al. 2022), the dataset is introduced as such:

"Movement sensor data from seven static and dynamic dog behaviors (sitting, standing, lyingdown, trotting, walking, playing, and (treat) searching i.e. sniffing) was collected from 45 middle to large sized dogs with six degree-of-freedom movement sensors attached to the collar and the harness."

Furthermore:

"The behaviors were annotated post-hoc based on the video recordings made with two camcorders during the tests (...) The annotated data was originally used for training behavior classification machine learning algorithms for classifying the seven behaviors."

And not only the dataset has been made available by the authors, but also the signal processing pipelines used for the classificaiton tasks, whose results have been published in another paper (Kumpulainen et al. 2021). This has sparked the idea of re-producing their analysis in order to better understand the dataset structure, before moving towards an edge approach. And this will most likely be the topic of our next labnote.

Not only that, the provided description of the dataset has also revealed both potential challenges and improvements in the original scope of our research. For instance,

"it should be noted that the analysis algorithms are fitted for middle to large sized dogs, which should be taken into account in the further application of the analysis algorithms."

Considering that the small to mid-sized Maniçoba has been our canine partner in this endeavor so far, this is definitely something that we must be aware of. On the other hand, the fact that sniffing is one of their labelled behavior provides an exciting avenue of exploration given that we're working with a AI enabled scent sensor in our research. Namely, the BM688 (embedded in the Nicla SenseME form factor):

One possible consequence of this is that whenever the edge-running behavior classifier detects that Maniçoba is sniffing something, we can active the BME688 ir order to get insights into what he's possibly picking up with his (certainly more sensitive) nose. But, there are some technical challenges to be conquered here, like the running of two different classifiers in the same edge-device. In all cases, this is an exciting prospect.

Another thing that called our attention in the "Description of movement sensor dataset for dog behavior classification" (Vehkaoja et al. 2022) is that

"other behaviors, such as drinking and shaking were annotated and are available in the dataset but those have not been used in the studies reported"

This represents an opportunity to expand on the original research the authors, as well as enrich our own sensory perception of our canine companions moving foward.

So in sum we're very excited with the opportunites (and the challenges) ahead and eager to share our progress with all of you.

Once more, thank you very much and stay tunned!