I am planning to make a low cost autonomous robot capable of navigating by itself. The robot will have a Raspberry Pi 2 B+ on board for processing and if needed, a laptop connected via WiFi placed elsewhere. It will run ROS (Robot Operating System) for ease of operation.
For sensors, I don’t want to use LiDAR or RGBD cameras like Kinect or Asus Xtion to keep the cost down. My plan is to use a webcam (or maybe 2 for stereo vision), wheel odometery data and a IMU sensor if required.
The first step is obviously to use SLAM and create a map of the environment. I tested ORB SLAM 2 with a monocular camera, https://www.google.com/url?sa=t&source=web&rct=j&url=https://github.com/raulmur/ORB_SLAM2&ved=2ahUKEwiszaem_vDhAhXRiHAKHUZtAB4QFjAAegQIAhAB&usg=AOvVaw3l_-DMtr4_YVqyTXmThvEs. It didn’t work with plain walls, needed visual variations for mapping and often lost track. I feel augmenting wheel odometery data can certainly increase the performance. Another thing could be to use some visual markers so that the camera can use them as landmarks, I would prefer not to use them but am willing to if needed.
I feel someone must have tried making some low cost navigation system for a mobile robot. Please let me know if you’re aware of any such project which is already implemented. Code, webpage, research paper, anything would be helpful.
Thanks in advance