This week I took some time to work with Ishaan on the software aspects of the project as well as worked with the entire team on the construction of the robot.
I worked with the team on the hardware this Friday. I was able to finish cutting the parts for the robot. However, as we assembled it, we realized that some of the cuts were really difficult and hard to assemble. While we were able to piece some parts of the structure together, after a severed screw, we decided that it would be best to redesign these with simpler cuts.
In terms of progress with the software, Ishaan and I were unable to integrate the computer vision code written in c++ with the robot motion algorithm that was written in python. To resolve this, we spent some time rewriting the code from c++ to python. This was a fairly easy process as it mainly only involved converting the code and not changing the algorithm. Additionally, we discovered the pycreate2 library. This library has wrappers for the open interface for the iRobot create. We were able to convert some of the more complex motion commands that I had previously written to simple use wrappers from this library. One such example is the safe wrapper. A simple call to this function makes it so that the robot drives in safe mode. What this means is that the robot will automatically halt if it feels like its going to crash into a wall. This can come in super handy when the robot is near the net or the walls of the tennis court.
Ishaan and I were ultimately able to get the robot to move towards a tennis ball! We have a base algorithm working where if the tennis ball exists in the camera view, then the robot will move towards the ball successfully. If the position of the tennis ball changes in between, then the robot will update its direction and follow the tennis ball.
In terms of my work for this upcoming week, I plan on making the algorithm more robust and better. Some this I plan on looking into this week are:
- If there is no ball in the camera view, the robot needs to turn and search for a ball
- Need to identify a good interval to do this check at since we can’t have the robot spinning aimlessly.
- Start using the depth feature of the camera so that the algorithm can work for multiple balls. The simplest algorithm that I can do is:
- Use the coordinates of the object and choose to go towards one with the smallest Euclidean distance.
- If there are too many balls on the screen, then do this process for maybe 5 balls and pick one and go towards it.
- Will doing this operation be too slow? How can this be mitigated?
- Test this algorithm somewhere that is not my living room 🙂