This project proposes a solution for position estimation of a single industrial manipulator using learning methods.
The model uses Region Based Convolutional Neural Network (R-CNN) to track robot keypoints which allows to predict the pose of robotic manipulator. The keypoints are defined as:
- A robot joint;
- The joint's "head";
- The joint's "tail".
The model prediction result examples: |
---|
Dataset was created and preprocessed using Roboflow.
Available in a Keypoints.v2i.yolov5pytorch.zip
archive.
Example of labeled image from dataset |
---|
The model is created, saved, and tested in the Pose_estimation.ipynb
.
Jupiter Notebook Demonstration.ipynb
is used for demonstration purposes, where both the dataset and model are loaded in to illustrate the predictions.
python v3
is required (v3.7+). Its installation guide can be found here- PyTorch Vision library for object detection The necessery files can be obtained from the repository directly.
- Additionally, the following python libraries are required:
torch
torchvision
albumentations
cython
pycocotools
matplotlib
numpy
These can be installed using pip
- Clone the repository:
git clone https://github.com/mirnanoukari/Pose-Estimation.git
- Run
Pose_estimation.ipynb
to create and save the model; - Run
Demonstration.ipynb
.