Decision Log
| Decision # | Date | People Involved | Topic | Decision | Alternative | Rationale |
|---|---|---|---|---|---|---|
| 1 | 2/2/23 | Alec Trela & Solomon Fenton | End-Effector Testing | – Testing End-Effector in a qualitative manner | – Look to link sensor data to gripping performance (force, torque) | – Too hard to link sensor data to performance in this setting |
| 2 | 2/20/23 | All Team Members | xArm is broken | – Move forward with the Sawyer arm for SVD and confirm availability with Kantor – Simultaneously, get a simulation environment working with an “arm-agnostic” planner able to handle both 6 and 7 DOF arms | – Entirely Simulation environment for the spring – Get some to buy an arm – Reach out to other resources (MFI) | – Discussion with sponsor – Sawyer seems like the only available arm this semester but will not be the arm going to Iowa so we also need to develop an “arm-agnostic” simulation |
| 3 | 2/27/23 | Alec and Solomon | Arduino vs Raspberry PI | Arduino | Raspberry PI | – Dynamixel shield compatible with Arduino – Didn’t have RPi |
| 4 | 3/11/23 | Alec and Solomon | Serial Communication Protocol | Use Arduino Mega and alternative RX/TX pins with FTDI serial to usb converter | Use Arduino Uno and SoftwareSerial to convert digital pins to UART pins | Arduino Uno route did not work, using the Mega is more reliable. Also leaves additional UART pins for future proofing |
| 5 | 3/15/23 | Alec and Solomon | Dual blades for cutter | Blades on both sides of the cutting mechanism | Single blade | – More effective cutting |
| 6 | 4/3/23 | All Team Members | Switch to Kinova arm | Use the Kinova Gen3 6 DOF arm in the Tepper AI Makerspace for the SVD | – Motion planning in simulation – Sawyer arm | – The Sawyer arm is not capable of handling the orientation constraints and obstacles necessary for our SVD |
Issues Log
| Issue Number | Date Found | Date Fixed | Change Made By | Origin | Issue Description | Resolution Plan | Artifact(s) | Resolved? |
|---|---|---|---|---|---|---|---|---|
| 1 | 1/29/2023 | 1/29/2023 | jiyoonp | No data on the internet | Don’t have pepper dataset | Made them from youtube farm videos | Dataset | Yes |
| 2 | 2/7/2023 | 2/8/2023 | jiyoonp | The format we created/labeled the datasets were wrong | AssertionError: Expected target boxes to be a tensor of shape [N, 4], got torch.Size([0]) | I changed the datasets that were being used to train the model. Like reducing the size from 1500 to 500 images. I assume there was something wrong with the datasets. (but when I checked for the images that were throwing errors, they was nothing wrong.) | Code | Yes |
| 3 | 2/8/2023 | 2/8/2023 | jiyoonp | The way imshow takes in arguments | normalizing the image and doing imshow will show me a black image | Don’t pass in normalized image → de-normalize and show an image | Code | Postponed as we are not using Faster R-CNN |
| 4 | 2/9/2023 | 2/12/2023 | shriishs | Colab code was set up but model would not begin the training | YOLOv7 will not start training | Speak to seniors and read official documentation to check for potential mistakes in code | Cleaned code and changed hyperparameters | Yes |
| 5 | 2/10/2023 | 2/19/2023 | sfenton | Cutter testing | Cutting results poorer than expected | Will need to experiment with other cutting mechanisms and higher torque motors. | Cutter prototype | Yes |
| 6 | 2/13/2023 | 2/17/2023 | shriishs | Model reaches maximum accuracy level and saturates | YOLOv7 validation accuracy saturates | Read online about hyperparameter tuning | Realized that the accuracy being acheived was state-of-the-art values | Yes |
| 7 | 2/13/2023 | 2/15/2023 | atrela | Testing of Gear Driven Cutter Design | Dynamixel SDK wasn’t driving motor | Ask Jash how to fix it and use the GUI in the meantime | Replaced Raspberry Pi for Arduino + Dynamixel Shield | Yes |
| 8 | 2/16/2023 | 2/23/2023 | sridevik | Meeting with Dominic Guri | xArm is not going to be easy to fix or is unfixable | email John Dolan, George Kantor, and Kroemer — see what they have to say and look for potential replacement arm options | Manipulator hardware platform, SVD testbed | Ongoing |
| 9 | 2/21/2023 | 2/23/2023 | atrela | Testing a Daisy Chain operation of our motors | Could not operate the AX-12 Dynamixel Motor with the Shield | Try refactored code, will try to run this past Jash as well if the refactored code does not solve it. | Swapped the AX-12 for MX-28 | Yes |
| 10 | 4/1/23 | 4/3/23 | sridevik | Testing with constraints on the Sawyer arm | The Sawyer arm’s reachability map is not good enough to handle our set of orientation constraints and obstacles | Switch to the Kinova arm in the Tepper AI Makerspace | Kinova Gen3 6 DOF Arm | Yes |
| 11 | 4/10/23 | 4/12/23 | shriishs | Full system integration led to inaccurate POI computations | Camera was not properly calibrated | Created a formalized camera calibration process using an Aruco marker, the Kinova camera, and the RealSense camera | Aruco marker, Kinova camera, code, calibrated transform values | Yes |
| 12 | 4/12/23 | — | sridevik | Switching from Kinova simulation to real arm | Cartesian moving with Kinova arm | Set longer durations on the movements so that the acceleration/velocity limits were not exceeded | Code | Ongoing |
| 13 | 4/26/23 | — | sfenton | SVD Encore | Dynamixel Motor Failure | Look into potential reasons for the motor failure over the summer | Dynamixel motors | Ongoing |