What it does
Our goal is to create a device with the aim of identifying recyclable and non-recyclable materials and minimizing the contamination in a recycling load.
Rice University, like many other universities, companies, and individuals across the country, recycles different materials to help protect the environment. In addition to the benefits of being green, Rice receives payment for the recycling stream that they deliver to the local waste management facility. However, if a significant portion of the recycling stream is made up of non-recyclable items, or contaminants, the load can be rejected, resulting in loss of payment and a fine. Our device will remove potential contaminants from the recycling stream, which will allow Rice to be more sustainable by reducing the amount of loads that are rejected
How it works
The system program would be run on the Raspberry Pi. The system would continuously poll the limit switch that was connected to the Raspberry Pi’s GPIO. Once it polled that the intake flap was open, the system would wait a second and check if it was closed again. The system would then communicate with a secondary Arduino to lock the intake flap. Once the intake flap is locked the system would then take an image with the camera attached to the Raspberry Pi and feed it into the onboard classification algorithm. Once the image was classified into either recycling or trash, the system would then send the result to the secondary Arduino which also controls the sorting mechanism motors. The system would then wait for the sorting mechanism to complete. Once it has communicated that the sorting is completed, the system will send a signal to the Arduino again to unlock the intake flap. Once the intake flap is unlocked, the system will reset itself to its initial state.
After developing an initial proof of concept to test sensors and collect image data, the team was ready to begin developing a full prototype. Due to some shortened timeframes, we set out to go from idea generation to a fully realized device in only a few weeks, and were able to accomplish it successfully. The design process consisted of 2 weeks in which the design was broken up into 4 subsystems, with 2 subsystems a week. The team then split to have an even number of members covering each subsystem, with both groups convening twice a week. Ideas were generated in the first half of the week, and then one idea was selected by the first all team meeting. In the all team meeting, the two subsystem ideas were discussed and compared to verify feasibility and compatibility. In the second half of the week, the possible implementations were considered, and the best implementation was chosen to begin determining parts and detailed specifications. In the second all team meeting of the week, last concerns regarding the implementations were addressed, and details of integration and ordering were discussed. This process allowed us to quickly iterate to a full design in a short period of time with a small team.
How it is different
While most technologies we found were either papers or hackathons, one competitor stood out as both technologically and economically viable - CleanRobotics’s Trashbot. While this device excels at buyer metrics and has begun to establish a market, there are some major weaknesses that this product fails to address. While the device’s set it and forget it mentality is certainly convenient, it fails to address a major problem of recycling education by giving no feedback or incentive to the user to recycle correctly, and perhaps even making their habits worse. By failing to educate the user, the contamination problem will continue to be a major concern. Lastly, since the device attempts to sort both trash and recycling, it again fails to address the main problems with recycling - contamination. The system has a high level of accuracy, but it is not sufficient to ensure the low contamination rates needed across recycling facilities.
There are many things to be worked on moving forward. For one, the algorithm will continue to be refined in order to maximize accuracy (while ensuring that the data is not overfit), and support additional categories and features (such as potential educational feedback or incorporating weight data properly). In addition, we will continue to expand our dataset.