karimhassanieh / karimhassanieh-udacity-robond-perceptionproject

Udacity Project 3 - Perception Project PR2 Pick&Place

Project: Perception Pick & Place

Rubric Points

Here I will consider the rubric points individually and describe how I addressed each point in my implementation.

---

Writeup / README

1. Provide a Writeup / README that includes all the rubric points and how you addressed each one. You can submit your writeup as markdown or pdf.

You're reading it!

Exercise 1, 2 and 3 pipeline implemented

Excercise 1,2,3 implemented in 'PR2_pickplace.py' python script which you can refer.

1. Complete Exercise 1 steps. Pipeline for filtering and RANSAC plane fitting implemented.

Implemented in the following order to the raw pointcloud data :

1-Statistical Outlier Filtering, with a set mean equal to 10 and standard deviation threshold equal to 0.001

2-Voxel grid downsampling with a leaf size equal to 0.01

3-A passthrough filter was implement, Z axis between 0.45 and 0.85, X axis between 0.33 and 0.9

4-RANSAC filtering was implemented with a maximuim distance of 0.01

2. Complete Exercise 2 steps: Pipeline including clustering for segmentation implemented.

Clustering was preformed with the following parameters taken into consideration :

Cluster Tolerance	Min Cluster Size	Max Cluster Size
0.05	50	200,000

The following images are the results obtained :

2. Complete Exercise 3 Steps. Features extracted and SVM trained. Object recognition implemented.

Features were extracted and trained using linear SVM model. 100 orientation were used to train the model (you may refer to capture_features.py and features.py for the code implentation ) . Below are the results obtained , the model had 83% accuracy :

Pick and Place Setup

1. For all three tabletop setups (test*.world), perform object recognition, then read in respective pick list (pick_list_*.yaml). Next construct the messages that would comprise a valid PickPlace request output them to .yaml format.

Message in yaml format are found in "output folder".

The robot sucessfully identified :

-3 out of 3 objects in world 1

-4 out of 5 objects in world 2 ( The robot kept mislabeling the book for soap)

-8 out of 8 objects in world 3

As a result the project was successfuly future work will include improving accuracy to fully recognize all object in world 2 and to complete the challenge (which unfortunately I could not complete due to lack of time )