For the week of September 20th – 26th, I focused on continuing my research and development into Iteration 1 of our 6-DOF robot arm. After initially choosing a YOLO model for AprilTag (used for calibration) detection, my team and I discussed using the NVIDIA Isaac ROS or OpenCV instead, since it is more optimized for AprilTag detection. However, since our design has not yet been fully finalized, much of our final electrical components have not been finalized yet as we do not have accurate enough data for torque calculations. In addition, we are still investigating whether a dedicated microcontroller will be necessary for our final system.
One of my main areas of research this week was determining how AprilTag tracking could be incorporated into our system. I investigated using OpenCV-based and Isaac ROS AprilTag detection approaches as alternatives to using a YOLO-based object-detection pipeline. OpenCV would potentially provide a simpler solution for identifying AprilTags, since the architecture is quite easy to use, which cannot be as easily said about Isaac ROS. This could simplify the computer-vision pipeline compared to training a YOLO model specifically for AprilTag detection. However, I am still evaluating the tradeoffs between the approaches, specifically in terms of detection accuracy, implementation ability, and how well each method would integrate with the NVIDIA Jetson Orin Nano and Intel RealSense Depth Camera D455.
I also continued investigating the smart bus serial servo motors that we discussed last week. While Hiwonder, Feetech, and Waveshare are still candidates, I was unable to definitively select the motors because our team has not yet finalized the Iteration 1 design, and as a result, we do not have an accurate estimation for the required torque at each joint. Therefore, we decided to hold off for the time being to confirm our final Iteration 1 design, since prematurely ordering motors that are too weak to support the joints is a real concern. In the meantime, I will continue searching more approved vendors such as DigiKey to source smart bus serial motors from.
Overall, this week was primarily focused on narrowing down the computer-vision and embedded-system architecture while continuing to work with my team on the overall design. Since our Iteration 1 design is still a work in progress, my work has consisted more heavily of research and comparison instead of final component selection. Once the mechanical design is finalized, I will be able to use the resulting simulation data to make appropriate decisions about the servo motors, power requirements, and camera placement.
For next week, I aim to complete the following:
- Continue comparing OpenCV AprilTag detection against YOLO-based detection and determine which approach is better suited for our system
- Test or prototype AprilTag tracking using the NVIDIA Jetson and camera setup.
- Collaborate with the team to finalize the Iteration 1 mechanical and electrical design.
- Use updated simulation data to determine the torque requirements for each smart bus serial servo.
- Narrow down and finalize the specific smart bus serial servo motors for each joint.
- Assist the team with refining the master parts list, including component prices, links, specifications, and quantities.