FtcRobotController VS ShellyVision

Yes, APK from Github, or just connect your phone to the REV Hardware Client.

I think I just stumbled upon where the team got the IMU initialization from. Take a look at the sample code in https://github.com/FIRST-Tech-Challenge/FtcRobotController/blob/master/FtcRobotController/src/main/java/org/firstinspires/ftc/robotcontroller/external/samples/SensorBNO055IMU.java

You can read my testing notes here https://github.com/FIRST-Tech-Challenge/FtcRobotController/issues/317 but there is a way to make it slightly faster

Although written for Blocks, this tutorial can help new Java programmers too: https://github.com/FIRST-Tech-Challenge/FtcRobotController/wiki/Timers-in-FTC-Blocks

are you working from the FtcRobotController project? https://github.com/FIRST-Tech-Challenge/FtcRobotController

Here's a sample program that uses deadlines

https://github.com/FIRST-Tech-Challenge/FtcRobotController/wiki/Writing-an-I2C-Driver This is the core tutorial on i2c devices, although its a little out of date so YMMV. Some things like the constructor may need to be tweaked (I believe there is now a second parameter in the super?). However the core of it is still accurate.

This might be a little late but here's a repo my team made to easily train TFLITE models for your robot.

For another option, if you'd like to use your own hardware, this is what our team used to train our model

The machine learning toolkit is great for beginners but is annoying in that you only have a specific amount of training time, training is slow on external hardware, and only coaches have any real access to uploading and such. Plus, even creating datasets with the tool is hit or miss as it creates some strange errors in training on your own hardware. I've uploaded the code we used to train our model to a half-baked repository on github. Hope it helps, or at least gives you another option