Conclusion

How well did your finished solution meet your design criteria?

We met the project goals we defined but were unable to meet our extensions due to time constraints on the robot. We intended the arm to sort multiple blocks and generalize our solution to more difficult problems, but due to limited Sawyer availability we were unable to fully develop and test extensions to our base project.

Difficulties

• Off-robot USB camera noise (white balance and saturation)
• Coordinate transforms between the camera and robot base frames
• Collision avoidance IK planning slow and/or not able to find a viable solutions
• Attempted switch to Baxter due to limited Sawyer arms
• Orientation of AR tags affected position calibration for picks (Potential reasons: gripper not factored in to move command, depth perceptions, base offset)
• Tuning TurtleBot mapping sensitivity to avoid falsely detected walls
• TurtleBot programs had difficulty planning twice in a row if something moved in between, causing random quits

Flaws and Hacks

Because we were using a off-robot USB camera, the position and the orientation of the AR tags relative to the robot base were off by various noisy amounts. Thus, we had to calibrate and apply various static transforms to each AR tag location. One potential reason for this could be that the robot limb did not register the gripper length in its position calibrations. The x and y offsets were usually very small and required only slight tuning.

Potential Improvements

• Remove the AR Tags and use computer vision to identify objects and the TurtleBot.
• Add machine learning or specialized path planning to enable the arm to take more direct paths to the goal.
• Feedback loop to reinforce Sawyer arm movement choices to improve consistency further.
• Dynamically allow Sawyer to plan around non-static obstacles.
• Enable the Sawyer to grip difficult to grasp objects that may require certain orientations or grip angles.

Image Gallery