The primary risks associated with our project right now are final integration and the robot’s ease of access when collecting a trash component while moving. During testing, we found that the robot has difficulty picking up the components while moving – due to the orientation of the roller & the speed. We are using an ultrasonic sensor to detect the trash at a certain location before coming to a stop to manage this risk. Once the component is within a certain distance, the robot will stop and begin the roller to collect the component.

A few minor changes were made to the design – the camera has been moved a bit lower on the robot and angled downwards to detect trash components better and the addition of two ultrasonic sensors (one near the front of the robot and one near the storage area – to note which components have been collected). We discussed the updated schedule in the final presentation, which was also delivered this week. The slide is given below. Our primary focus for the next week and the days leading up to demo are fine-tuning integration parameters, final testing to ensure functionality & consistency, and documentation (final report, poster, and video).

 

List all unit tests and overall system test carried out for experimentation of the system. List any findings and design changes made from your analysis of test results and other data obtained from the experimentation.

Multiple tests were run on the robot to test functionality and durability. Tests were also run to check whether the robot meets the design parameters and use case requirements.

  1. Movement:
    • Testing for this subsystem consisted of whether the robot is able to move forward and rotate (to look for other components). From iteration testing with the robot, we found that the robot is able to move forward and rotate. A video of more recent testing (pick up + movement) is given below.
    • Moving
  2. Conveyor Belt + Roller (pick up mechanism):
    • Testing for this subsystem consisted of whether the system is able to collect a trash component or not. Time data (how long it takes for the robot to collect a trash component from detection to pick up) is collected during integration testing.
    • We found that, when stationary, the robot is able to pick up trash components. A video is given below. 
    • Stationary Collection
  3. Object Classification:
    • Testing for this subsystem was done earlier on in the semester. The results have been discussed in detail in the final presentation as well as the interim demo. The numerical results are provided below:
      • ~115 FPS for inferencing, ~103 FPS for full ML pipeline
      • 2 & 3: mAP = 0.978 avg. across classes @ 0.68 conf. threshold
      • 2 & 4: Recall = 0.975 avg. across classes @ 0.68 conf. threshold
  4. Integration 1 (movement + collection):
    • Results from this testing are still ongoing. This week we integrated the ultrasonic sensor at the front of the robot. We will be collecting test data (time it takes for robot to reach & collect trash item and the number of items it collects in total) for this section before the beginning of next week.
  5. Integration 2 (detection + movement):
    • This test was mostly to check whether the robot can act on/ respond to results from the camera. Quantitative test data was not collected, but we saw that the robot is able to move in response to an object being detected/ not being detected.
    • The robot has shown that it can use results/ data from the Jetson Orin Nano & follow the directions given by the pi (move towards object if trash component has been detected otherwise begin rest of flow – search for next component).

0 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *