RL-EMPC-RampMetering

Using Reinforcement learning (RL) based event-triggered Model predictive control (EMPC) to solve the ramp metering problem.

File Explain

src: core code

agent: RL agent

mpc: mpc controller

opt: optimization controller

simulation: model and env

docs: related documents

tests: test code

examples: example code

Track the Deer

This repository is a part of the research : A Comprehensive Evaluation of YOLO-based Deer Detection Performance on Edge Devices

This github repository contains the source code used for training and evaluation of YOLO models, exporting them to ONNX and using the ONNX models for inference testing on edge devices; NVIDIA Jetson AGX Xavier and Raspberry Pi 5.

Directory Tree (Main Structure)

track-the-deer/
├── cfg/                # Configuration files for training and datasets
├── data/               # Datasets and raw data
├── experiments/        # Training outputs and logs
├── inference/          # Inference scripts for different devices
├── notebooks/          # Jupyter notebooks and analysis utilities
├── onnx_models_op13/   # Exported ONNX models
├── output_images/      # Images generated from inference and annotation
├── runs/               # Detection results
├── src/                # Source code for training and testing
├── requirements.txt    # Python dependencies
├── Makefile            # Build and automation commands
└── README.md           # Project documentation

NOTES

Dataset

For dataset, refer to https://www.kaggle.com/datasets/winnerbishal/deer-cameratraps .

How To

The structure streamlines the training process. For instance, models can be trained with ease using commands like make run CFG=cfg/run_Lila.v1.3k.yaml . The content of cfg/run_Lila.v1.3k.yaml is as follows :

model: yolov9s.pt 
data: cfg/Lila.v1.3k.data.yaml
epochs: 100
batch: 48
imgsz: 640
workers: 32
project: experiments
name: Lila.v1.3k_y9c_ep100

After training, make run export_onnx to generate ONNX models.

Inference in NVIDIA Jetson AGX Xavier : jetson_cuda_inference.py

Inference in Raspberry Pi : pi_inference.py

Inference is performed on image batches while recording the predictions, inference, pre-processing and post-processing times, and other performance metrics. Finally, a csv file is returned.