‍Through desk research, academic research and interviews with the NGO, we mapped out different types of reed eliminating methods:

We opted for over-surface on-water cutting considering 2 factors:
1) on-water
It is the most tedious for humans from the above-mentioned methods.
2) over-surface
Although it would be most effective to cut the reed as close to the root as possible, it would disproportionately increase the difficulty of maintenance and costs. Over-surface cutting slows down the growth of the reed enough so that other plants have time to strengthen.
3) spouts
It is a decision made later in the process. For more, see "Researching the Lifecycle of Reed".

We explored different cutting mechanisms.

Initially, we prototyped the simplest one which was a pair of blades attached to a servo motor in a scissors-like way (or in our case, an actual pair of scissors). We planned to have it as our first iteration. However, we conducted performance test by cutting actual reed with it and surprisingly, it proved to be sufficiently powerful to cut through the stem. Therefore, we decided to stick to this mechanism.

Our two main goals were to develop a cutting mechanism and a plant-recognition system. To design the latter, our initial approach was creating a vision-based system. However, we realized that that was disproportionally energy-consuming which is an important factor in case of a robot that functions in nature. Therefore, we decided to deviate from a “human-perception-based” approach (identifying reed by visually recognizing it) and explore low-level sensors. Through  researching different sensors and testing, we found that a capacitive sensor is capable of distinguishing different materials. Although it is not precise enough to base the complete reed-recognition on only capacitance, it serves greatly for a continuous low-level monitoring of the environment, as well as for distinguishing living plants from other entities of the ecosystem.

In the previous phase, while exploring different sensors, we also learned the traits by which reeds can be identified (leaf width, legule, etc.). But more importantly, we learned about its lifecycle. We first emphasized the importance of NOT cutting the reed in the period when it releases its spores. However, after revising the interviews made with members of the NGO, we noticed a point that we previously overlooked: the interviewee recommended that the robot cuts the spouts instead of fully developed reed. Previously, we didn’t tackle it as our focus was on easing the physical burden on humans. Cutting the sprouts was not done by them since they grew over the water at a different pace, so it was more practical to wait until they all grew completely. Once we prioritized the nature, eliminating the sprouts seemed to be the best to prevent reed-invasion since in that case the reed simply has no resources to maintain itself. This method eliminates the reed instead of just slowing down its growth. Thus, interestingly, it is assumed to ease the burden on humans since the robot would do the first-round elimination of the reed, instead of attempting to replace humans by doing the same job as they do.

Upon realizing that the capacitive sensor is capable of distinguishing non-alive entities, plants, and animals, but not precise enough to distinguish reeds from other plants, we brought back the visual recognition idea. We got started with the idea of using an API, however, it can be problematic since a stable internet connection can not be taken for granted in wetlands. Instead, we propose the concept of an AI specifically trained to distinguish common reeds from the other members of wetland vegetation. In our prototype, we used image recognition with a Teachable Machines Machine Learning Model.
‍
(NOTE: in the GIF, you can see that we connect the motors to a phone's camera for the scope of prototyping)

In our first step (Narrowing Down the Framing), we set the goal of exploring 3 areas in the development of Reed-E:
1) the cutting mechanism
2) the plant recognition, and
3) how to combine them.
We initially added a pair of grabbing arms to assist the cutting mechanism. In the meantime, we experimented with capacitive sensors.
Interestingly, we found that we could enhance the grabbing arms' functionality by adding the capacitive sensor. With this step, we realized our 3rd goal: the grabbing arm helps both the plant-recognition by providing the 1st-step identification & the cutting mechanism by stabilizing the stem.

+

=