This week, Hojun, Abha, and I spent some time figuring out how we can connect to the jetson in Hojun’s room over the internet. After many trials and errors, we finally got it to work.
I also was able to rig together our depth sensor and IMU on an RC car to simulate how our RTABMAP SLAM algorithm would work in real life without actually having the physical robot. The algorithm seems to be overkill and computationally expensive for our needs but has shown to be a good tool to create a solid map for our surroundings. It also requires a lot of tuning. As a result I have chosen to pursue an ORB SLAM method of mapping and localization which is an extremely accurate, fast, robust monocular SLAM method. I started to try to setup ORB SLAM2 in ROS which uses stereo vision instead of monocular vision which should provide better results than the original ORB SLAM. In conjunction, I plan to use the gmapping ROS package which allows me to take stereo images and create a 2D occupancy map. I was able to get gmapping working with the simulated turtlebot3 robot and in real life but the real life results are more noisy than the turtlebot3 results.
I also assisted Andy in our color thresholding object localization.
Next week, I plan on integrating our software into the jetson with Hojun and running SLAM on the robot.