Assess the performance of the object detection pipeline for a single object category (bottles) under controlled conditions.
| Associated Requirements | Untitled (https://heathered-english-a1d.notion.site/cfa7d0cbbc7c493db42afd2419737e1a) |
|---|---|
| Equipment | Stretch RE1 |
| Elements | |
| Personnel | Lead: |
| Location | AI Makerspace, Tepper School of Business |
a) Place the robot at a fixed location in front of the table.
b) Mark 5 locations on the table (using coloured tape) where a given object can be placed. Ensure that there is sufficient lighting on the table.
c) Place 5 different bottles of varying shapes and sizes on each of the locations of interest.
Turn on the robot, home it and enable the camera node.
Using the robot, collect a dataset of images using the "rosbag" tool that is supplied with ROS. The dataset should be collected as follows:
a) With a given setup of 5 bottles in their locations, run the rosbag for 5 seconds.
b) Interchange the positions of the bottles, and repeat "3. a)". This is done 3 times.
Using the collected dataset, mark the bounding boxes on the actual location of the object in the image on the dataset. This will serve as the ground truth.
Run the object detection algorithm on the collected rosbag, and compute the mAP with respect to the annotated ground truth collected in the previous step.
Kill the script and turn the robot off.
The computed mAP is >= 80% for different kinds of bottles in each 5 second interval.
Date of Completion: February 28, 2023