penny-pi

a small self driving robot that loves our cat named penny :)

---

in action

current features

Feature	State
Bluetooth Driving	✅
Self Driving	✅
Obstacle Avoidance	❌
Cat Detection	🟡
Tag Following	🟡

hardware

none of the links work yet

• chasis: todo
• "brain": raspberry pi 2
• camera: raspicam
• controller: l298n
• lights: generic GPIO lights
• speakers: generic GPIO speakers
• bluetooth: generic bluetooth USB

software

currently, we can just connect to it with SSH, after that, we just do the follwing:

#wherever penny-pi is located
python penny-pi/server.py

#now in a different terminal/tab
sudo rfcomm watch hci0

then, you are able to connect on a bluetooth pairing app. we just use one found on the app store, and to adjust the input values (ex: "A"-> go forward, "B"-> go backwards), you can just adjust the very top of the server.py file (not implemented yet, you must scroll down in the code to do this)

machine learning

we gathered roughly 5k images total split evenly between the three classes ['F', 'L', 'R'] from the camera + normal bluetooth driving. once we recorded the data, we did a quick analysis of it through npz-reader.py, then ran it through a simple CNN. we got roughly 79%, but it suprisingly worked really well.

the next part was pretty basic, you can find some of our models in /machine-learning/. once we train one, we then convert it into a .tflite file so it can run on the raspberry pi. afterwards, we just used the documentation to making a simple set up of inputing the pixel array from the image, and outputting what the model wants to do.

#predictions will print in the `python penny-pi/server.py` terminal like below
[0, 1, 0] <- go left
[1, 0, 0] <- go forwards
[0, 0, 1] <- go right
...