mikaelfangel / golfbot Goto Github PK

• Handle if the robot's main crash, and the client should reconnect
• Make robot stop completely, if everything goes bad.
• Handle if the client crashes.
• Handle if we lose connection to robot.

• Make the rotation precise and reliable
• Make the driving straight precise and reliable

Make the balls categorized on where they are placed on the course.

Its valuable for us to make our code more reusable, so that when we are to make more complex calculations, the basic ones are easier to use.

• Detect the distance from the robot to a golf ball.

• Find markers and attach to robot
• Test precision and distortion effects
• Get angle from normal coordinates
• Run robot, with difference in angles
• Test if the robot has issues rotating on the course

I have an issue using the course object, because for each frame a new Robot object is instantiated, and I therefore cannot use the Robot object to stream. I tried to solve this, by only having a single Robot object, but would resolve in error on the vision-part. There is different robot objects store in both detectionController and robotDetector

• Make the robot drive forwards
• Make the robot drive backwards
• Make the robot drive at adjustable speeds
• Make the robot stop
• Implement individual adjustments
• Add error handling
• Write tests

We need to make functions to drive the robot, instead of hardcoding RPC requests.

• Make Manger Class
• Move default speed to Client

Improvement to vision from last iteration. Better accuracy in detection to balls, courseframe and cross.

• Detect orange ball with hsv filter
• Add hsv filter to better detect white balls -> ended up doing binary threshold instead because of the orange ball
• Explore use of findContours vs HoughCircles -> maybe later, not necessary atm, because the threshold filter improved reliability alot
• How does the robot affect the detection -> We need to cover white colors of the robot to not false detect
• Better detection middle cross -> waiting to solve this for later
• Explore camera distortion -> not needed at the moment

• Research opencv
• Detect 1 or multiple balls
• Detect course frame
• Calculate coordinate system
• Test precision and camera distortion effects

We might need to look into camera distortion to solve the mapping of coordinaters from pixels to irl measurements

We need to make sure that there is next to little loss in precision from what the camera sees to how the course is really layed out.

• (DROPPED) Correct camera lens distortion, if needed
• Map coordinate system to course, instead of camera frame

• Create design diagram

• Clarify which motors you can call.
• Specify ports on the ev3

The robot should be able to drive to the small / big goal and angle the delivery into the goals.

• Robot can rotate based on an input based on degrees
• The gRPC protocol includes the interface for the function
• Implement error handling

We need to convert our vision algorithms to java and collect everything we have made to a single file or some other single structure.

We need to adjust the color threshold and make it more stable.
Additionally we need to make us able to tune the parameters at runtime in the setup phase.

Make a method to either send multiple commands that is queue or let the client know when a function has finished. This should allow us to execute subsequent commands immediately.

• Locate Mark from video feed
• Get rotation from marker
• Convert rotation to usable units

• Count and store number of balls in the robot
• Make Balls Consistent using window median.

• Fix robot not stopping when driving to a point

• Create initial draft
• Write the protocol as a protobuf file

• Make RPC Call
• Tweak variables to fit to real world actions
• Test

• Research optimization of streaming, so we can stream more messages pr second
• Fine tune driving
• Fine tune rotation and possible also use streaming to turn accurately
• Address high coupling on client side