Instructions

1. Download this repository.

2. You need to put the following file inside the "Files" folder:

   https://drive.google.com/file/d/1s02V8NMhNPtBrqPgOX9J0mI5c-nGsmkz/view?usp=sharing

3. You need to have a pretrained mlagents model inside the "Files" folder.

   By Default, there's a model pretrained by our Master Student George.

   To know how to get mlagents to output a trained model to pytorch rather than ONNX, check the html file inside this repository.

4. Run the file you want.

File Description

There are 5 Main Files:

1. Debugging_Angle --> To visualize the camera stream while printing angle.

2. Debugging_Lidar --> To visualize lidar readings in real time.

3. Debugging_Stage --> To visualize the camera stream while printing stage.

4. Environment_With_Visualization --> Visualize the camera stream + RL Env to control robot.

5. Environment_No_Visualization --> RL Env to control robot (NO camera visualization)

HOW TO OUTPUT MLAGENTS TRAINED MODEL AS PYTORCH MODEL?

An explanation of How to modify the MLAgents library to output pytorch model is included as HTML.

You must have mlagents version 10 or above to be able to do this.

Credits:

1 RPLidar communication was achieved using this library:

-- https://github.com/Roboticia/RPLidar

2. Pytorch to TensorRT was achieved using this library:

-- https://github.com/NVIDIA-AI-IOT/torch2trt

3. In general, none of this would be possible without Victor Ham's help:

-- https://github.com/victorhamc

LIBRARIES USED IN THIS REPOSITORY

The libraries used to create this project were:

• CUDA==10.2.89
• jetbot==https://github.com/NVIDIA-AI-IOT/jetbot
• mlagents==0.22.0 (release version 10)
• opencv==4.1.1
• pillow==8.1.0
• pytorch==1.7.0
• rplidar==0.9.2
• torchvision==0.8.0
• torch2trt==0.1.0
• trt-pose==0.0.01

THIS IS A WORK IN PROGRESS

These files are wrappers for other libraries. Don't forget to give the authors credit for the hard work!