Sprint Goals

Sprint Goals

Sept 15

• ✔️ System profile evaluation code implement and try on QuadricSLAM and OA-SLAM
• ✔️ Clean scripts for automating the plot generation from the output of QuadricSLAM and OA-SLAM
• ✔️ Collect the BOP dataset generator from Sathwik and setup in my system.
• ✔️ Perform QuadricSLAM and OA-SLAM on a dataset generated.
• ✔️ Generate 5 datasets via cluster or my laptop (get access to the cluster)
• ✔️ Perform QuadricSLAM and OA-SLAM on the 5 datasets
• Update results in the report with plots as per the experiment results
• ✔️ OA-SLAM theory section in report complete
• OA-SLAM code explanation in report complete
• ✔️ Comparative Evaluation of QuadricSLAM and OA-SLAM - both based on implementation theory and evaluation results
• ✔️ Refine the report and share it with professor on Sept 03

June 07 - Mid term

1. ✔️ Quadric and QuadricSLAM summary create in the form of pdf from notebook.
2. ✔️ Make OA-SLAM working in local system.
3. Create block diagram of the working of OA-SLAM.
4. OA-SLAM paper summary.
5. ✔️ Compare the BOP Dataset estimated and ground truth of the objects by visualizing the pose of the object rather than quadrics in matplotlib.
6. ✔️ Metrics to compare the ground truth and estimated pose of the object.(cpu load, pose accuracy, pose error, run time)
7. Midterm report - comparison of quadricslam and OA-SLAM on BOP dataset.

April 15

1. ✔️ Fix the data assosciator part and make QuadricSLAM work on TUM dataset.
2. ✖️ Define what is quadrics, equations of a quadric and plot a quadric. what are the advantages and disadvantages of quadric?
3. ✔️ Understand the code and prepare Block diagram of workflow of the QuadricSLAM.
4. ✔️ Tutorial for understanding factor graphs in GTSAM for landmark-based SLAM.
5. ✔️ Make dataloader to load the BOP dataset and make it run on quadricslam.
6. ✖️ Brief note on the summary of the paper related with QuadricSLAM.

March 15

1. ✔️ Make QuadricSLAM work. - [Made sample example runnable]
2. ✖️ Check whether other detectors can be used in QuadricSLAM.
3. ✖️ Block diagram of the working principle of QuadricSLAM.
4. ✖️ Familiarize with the factor graphs.

TO DO

1. Create or collect world models that can be used in Gazebo for SLAM.
2. Familiarize with mapping algorithms both theoretically and ROS implementation - Gmapping, Hector SLAM, Google cartographer, Karto SLAM, RTAB map, Octo map.
3. Familiarize with localization algorithm both theoretically and ROS implementation - AMCL.
4. Create a ROS package with launch files to bring-up the environments(minimal environment and complex room) and robot model. That is, setting up the workflow such that each SLAM algorithm above can be tested on this common environment.
5. Also setup the workflow such that the algorithm can be tested on a rosbag file instead of a GAZEBO environment.

TO DO

1. Survey of papers related with challenges in AMCL and other later modified versions of AMCL.
2. Start survey of atleast 1 semantic SLAM algorithms.
3. Collect semantic SLAM algorithm implementations.
4. Try this algorithm on the GAZEBO environments created in previous sprint.
5. Identifying and comparing different ways of storing a map.
6. Also collect some public 3D dataset which can be used for testing the above algorithms.

TO DO

1. Identify the challenges with AMCL on 2D world.
2. Start creating challenging GAZEBO environments to test the limit of the 2D mapping and localization algorithms.
3. Identify the evaluation matrices that can be used for comparison of different algorithms.
4. Continue survey of 2 semantic SLAM algorithms and collecting their implementations.

TO DO

1. First draft of the mid term report.