Arka set up the RPi SSH capabilities and the camera pipeline. He tested and timed the object detection code on the RPi and started working on the resistor classification portion for CV. Claire set up the stepper motors controlled with the Arduino. There is more documentation on how to use these motor drivers with an Arduino than an RPi and the code is easier, so she is using this as a starting point to learn how to control the motors….
Schedule Per our Gantt Chart, this week I was supposed to set up the stepper motors. I was able to do some of it, but a lot of this week was focused on the Design Proposal. Progress This Week This week, besides working on the design proposal slides, presentation, and report, I was able to obtain the stepper motors and drivers and check whether or not I could interface them with an Arduino Uno. Using an Arduino Uno allows me…
This past week I focused on manufacturing the parts needed for the XY gantry and researching what parts to use for our Z-axis. I 3D printed the mounts but our rails still have not arrived yet so I could not move onto assembly like I planned to. We have decided to use an off-the-shelf part for our Z-axis to minimize the risk and time it would take to test a 3D printed rack-and-pinion part. We have also set up the…
This week I did a lot of research into part purchases with Maxine. I looked into the motors, and how to control the different parts of the mechanical system. We knew that we wanted 3 stepper motors to control the XY motion, some way of gripping, some way of rotating the gripper, and a way to control the Z motion (as mentioned last week). After research, we decided on a NEMA 17 with a A4988 stepper driver for ease of…
This week our team has been focused on parts setup and finalizing design plans, and preparing for our design review presentation. Arka has finalized the CV algorithm for object detection. He has been testing with images of breadboards and using those to tune the object detection algorithm. He has also been having some trouble setting up the RasPi on the CMU wifi, but will continue to look into it this weekend. Claire has been researching on the theory and power…
This week, I finished working on the object detection pipeline. Last week, I was struggling to properly filter our the background resistor board. I figured out the appropriate HSV values to filter it out, which should stay the same under these lighting conditions across images. I used CLAHE to reduce the effect of reflections of the resistors, morphological closing to close any holes that appeared in the resistors, and contouring to find and draw edges around the resistors. I cleaned…
After the research and planning we did last week, we put in the order for XY rails with the dimensions that would work with our project. I also spent the week finalizing the 3D printed components that we would need to mount our rails to our chassis, and mount the timing belt and servos. I printed the parts on Thursday night in Techspark, so hopefully when the rails arrive on Monday I will have all the parts ready to start…
This week our group has been focused on researching solution approaches to the two main components of our project: the robotic claw and the computer vision algorithm. We have also checked out some parts out of the 18500 inventory for some preliminary testing. Arka has started looking at image processing and deep learning techniques for the object detection part. He will first experiment with the openCV library’s image processing functions to see if they are accurate enough for our use…
This week I focused on researching different solution approaches for our robotic claw with Claire. In particular, we looked at some DIY claw machine projects powered by Arduinos and found them promising. We will be using stepper motors and an x-y gantry for the positioning of the claw. However, the up-down motion of these projects mostly used a pulley system that looks more flimsy than what we want, so we are now researching a different approach for that particular axis…
For this week, I focused on figuring out how to detect resistors on a breadboard and get their locations as accurately and efficiently as possible. I considered standard deep learning approaches that would find bounding boxes around resistors it found in a given image. The three alternatives for this were Faster RCNN –> the most accurate but slow and takes up memory YOLO –> much faster and lightweight alternative but accuracy is lower With both of these options as back…