PSMove can be connected to your Raspberry Pi via unofficial PSMove API: 

https://github.com/thp/psmoveapi

It was a bit tricky to install it on the Pi since it has a lot of dependencies.

For this application, Python 3 is necessary so make sure you have that installed first.

Then, follow instructions here:

https://psmoveapi.readthedocs.io/en/latest/build.html#building-on-ubuntu-18-04, also, make sure it generates bindings for Python 3: 

https://psmoveapi.readthedocs.io/en/latest/build.html#python-bindings 

cmake .. -DPSMOVE_PYTHON_VERSION=3

When you have the bindings working correctly (you can test by running demo apps included in PSMove API), you can proceed with the machine learning portion!

First, you'd need to install MXNet on your raspberry Pi. 

Here are the instructions: 

https://mxnet.apache.org/get_started?platform=devices&iot=raspberry-pi&

After you have MXNet working, download my repo

https://github.com/msurguy/MYHouse-smart-dollhouse

Try to run the main script by executing

sudo python3 myhouse-gesture-recognition.py

If that works, proceed with the connections of components!  

The connections to the Raspberry Pi pins are as follows:

• 6 = Fan servo
• 4 = Left shutter servo
• 22 = Right shutter servo
• 18 = Neopixel strip that consists of 29 LEDs that are split in two groups for letter M and letter Y

We made a data collection script located in https://github.com/msurguy/MYHouse-smart-dollhouse/blob/master/data-collection/collect.py

You can run it by executing

sudo python3 collect.py

When the data is collected, you need to train the network by executing this script: https://github.com/msurguy/MYHouse-smart-dollhouse/blob/master/training/Myhouse-training.py

python3 Myhouse-training.py

Then you can use the resulting network to operate as the brain of your smart doll house!