Agro-Bot 0001

AgroBot is the largest mechanical project I have ever undertaken. It is a story marked with struggle, an 8 week proposal-to-presentation turnaround, learning team project management by fire, and incredible amounts of work.

AgroBot began as a solution to the problem posed by the MassRobotics Form and Function competition— Create a robot whose form is unique to it's function. Initially, I had very few ideas about where to take the competition project, but with such an open ended project prompt, it had to be a massive feat of engineering to hold up to the competition.

Before ideating, I wanted to make sure that I had a team that would be able to stand up to the task. This ended up being one of the most challenging parts of the project, and I learned how difficult project management was. I went through the effort to put up ads in the Engineering buildings— Posters and flyers that I hung up through the walls with the massive text "'$10,000 PRIZE! SIGN UP TODAY!". I went as far as the engage the BU billboards outside of the second largest dorm building in America to display my branded poster, trolling for team members who would take the initiative to fill out my application and go through the team interview process. I held about a dozen interviews with prospective team members, and ended up with a team of six engineers across various years in their education.

These engineers consisted of myself, a Mechanical Robotics engineer, Grishen Hestiyas, a Computer Scientist whom I have known for many years, Howard Wang, another computer scientist and friend of Grishen, Nova MidWinter, an aerospace concentrated engineer with a specialty in systems routing design, Advaith Somula, a Master's ECE student with a focus in AI, and Vikram Bhalla, a sophomore with a hunger to learn. What I asked in these interviews (And what I valued the most from my team), was centered on the idea of showing up. Will you show up, and will you put in the work to make this project succeed? With emphatic agreement all around, I was feeling confident about choosing an ambitious project.

Brainstorming was not a very long process, because when we were initially thinking of ideas, Advaith suggested off the bat that we look into making an agriculture robot, as he had prior experience with the area. As a chef and food lover, I was immediately on board. It has been a long dream to be able to automate some amount of small scale, decentralized agriculture, and this was the perfect opportunity. We discussed which parts of the problem we wanted to attack, and more importantly, which parts of the problem would be reasonable to tackle. After deliberating for a while, we decided that a fruit picking robot would be an ambitious but reasonable goal given the team, resources, and prior experience. We did brief preliminary research for existing solutions, and found that the largest hurdles were the grippers, the computer vision systems across time, and the final price of the robots. Similar products were hundreds of thousands of dollars, and as a student team, this forced us to think outside the box.

Through the competition, we were given a pre-filtered list of sponsors to reach out to, but there were no previous arrangements for sourced parts. This led me to step into my first taste of project management role, taking up the fundraising job as the only person on the team with the full perspective of the robot we were building. I dove in for the first week into coordinating with the interested companies to engage them for as many components and parts to stick on our robot as they would allow. Many of the company contacts fell dead, because the representatives either went on vacation, or the project liaisons became too overloaded with the other teams to engage us. I was able to find agreements with a number of companies including Harmonic Drive, Copley Controls, Analog Devices, and Maxon Motors. The companies were all excited to get their devices and products into the hands of student teams, giving us heavy discounts and often sample products for free. Over the competition, I fundraised about $20,000 worth of components to build this robot.

When we had first gotten the project proposal approved, and attended the information session outlining the sponsors of the competition, I began immediately on designing the robot. The way I had understood it, I believed the robot's form being "unique" meant unorthodox. This ended up leading us down a difficult track, and ended in an important lesson about how current robot arm designs came to be. The CAD was initially meant to be split between me and the other Mechanical engineer, Nova, but when classes intensified, the work load was primarily placed on my shoulders. After 3 marathon nights working on the mechanical design for the robot, I had a completed CAD model that I was confident would provide a solid foundation for our plant-picking needs.

After working with the CAD model for a week or so, redesigning to accommodate the supplied sponsor parts, I was ready to begin fabricating the robot.

In the meantime, trying to juggle my responsibility as effectively the sole mechanical engineer designing the robot, and the project manager for every other department (Computer science and ECE), I found it nearly impossible to hold myself and everyone else accountable at the same time. My approach that I tried was to have weekly/biweekly meetings to discuss projects, goals, and progress. This worked for the first 2 meetings, but when trying to maintain the schedule through spring break and midterms, nearly half the team went radio silent as they took care of their academic affairs. This was a challenge that I anticipated when working with active students, but I underestimated how severely it would impact the project when the chips were down.

Through spring break, I was actively encouraging the electrical engineering side of the project and the computer science team to begin working early, because the sooner they could begin, the more time we would have to troubleshoot and trek through integration hell— A MASSIVE undertaking for a project of this scale. I found that with little experience ideating for robots, and an incomplete base of knowledge, a significant portion of the team did not know how to manage their part of the project, sitting idly until I spelled out the specific steps to complete their deliverables. Being a similar age and educational level (and even younger and earlier in my education than a team member), It was difficult to take authority and suggest that I knew better than the others, even when they blatantly said they did not know how to begin. This was the hardest part of the project management, because I assumed everyone was as equipped as me and that everyone has a complete understanding of all of the inner workings of the project like I had at that point.

Over spring break, I was putting an incredible amount of work into beginning the fabrication of the chassis of the robot. I had finalized (tentatively) the CAD, and had shipped the STEP model and the STL files to Advaith to create the URDF model so that we could begin Sim to Real teleoperation. This would go on to take another 3 weeks until he could produce a URDF model that we could import into CoppeliaSim and begin our motion planning. Fabrication was an incredibly time intensive process. During the week off, I spent 8 hours every day in the machine shop fabricating many of the ~160 components from the blanks of material I had purchased from out of pocket— That is to say there was not a lot of room for messing up and retrying. Being the first time I had actually utilized any kind of CNC machining for production (not just a demo), It was an incredibly steep learning curve to be able to go from CAD to native CAM, planning and implementing all of my toolpaths effectively on the fly. While I was machining one part, I was sitting in front of the E-STOP on my laptop CAM-ming the next component that would go in the machine.

After nearly 3 weeks of machining and 5 out of 8 weeks spent from our project budget already, the project became a frenzy of missing components, company negotiations due to short supplies, long shipping times, and other technical difficulties. The fabrication was taking place until the week before the competition, refitting odd pieces due to faulty CAD models, incorrect hardware being supplied, or a shortage of community parts like screws (Taken from the communal robotics lab). With nearly 14 of my 24 waking hours being spent already on the mechanical engineering Fabrication, CAD/CAM, and supplier negotiations, the other teams were falling behind quickly, and without the bandwidth to support them and run my own team, the divide became quickly apparent as I was nearing the end of the manufacturing phase of the project.

The electrical engineering team, who I presumed would be fine with project management, was completely inoperable. Advaith, the masters student, had a lot of experience in completing the technical aspects of the project such as basic PCB design and standard electrical engineering (his undergrad degree being in Electrical Engineering), but had very limited experience in setting timelines, deliverables, and planning out how to get from proposal to final deadline. This meant that while I was managing myself, The electrical team was waiting for more granular tasks to be sent their direction. I had initially told them the connections, given them all of the spec sheets, told them how the system was meant to connect, and requested that they give me a list of components that they were planning to use to get the deliverables finished. I did not notice that I had not yet received a list of components until it was far too late— A grave oversight on my behalf.

After taking nearly another week to respond, I had to cut Vikram from the team. I had multiple discussions with him, and we decided that because he was unable to devote any of his time, and the time he could devote was useless because he did not have the motivation to learn about the system we were building, it would be a better fit if he focused on his classes for the rest of the semester. This left effectively only Advaith working on the electrical engineering aspect of the project, and without clear direction or being easily contacted, he would disappear for days at a time with no coordination. This was all falling apart towards the end of my fabrication stretch, so It ended up falling on my shoulders to design the electrical system from power supply to motors. With only 3 weeks until the presentation, haste was paramount to the completion of the MVP. I spent 2 days creating the full electrical schematics to connect the 3 dozen components, including the communication data lines. After rush-ordering many of the components from Amazon (Prime shipping saved this project), the focus shifted from pure mechanical engineering to Electromechanical integration hell. I pulled 3 all nighters over the next week wiring, and ended up wiring nearly 500 feet of wire onto this robot that wasnt more than 2 feet in width. The communication lines were the most temperamental, because for the motor controllers to connect to the motor control boards, I had to solder 16 D-Sub terminal connections for 10 separate controllers. This took forever, and with cheap communal (low grade) wire causing multiple internal disconnects (the wire severed inside the insulation), troubleshooting was about 60% of the time budget.

This rushed timeline paired with fault components, sub-par wire, and a lack of sleep made those 3 weeks some of the hardest work I have ever put in. I was putting on average 100+ hours into this project each week, acting as the controls engineer, lead (and only) mechanical engineer, and only Electrical Engineer, the responsibilities of 4 of my 5 team members fell onto my shoulders. I do firmly believe that there could have been better project management practices that may have relieved a portion of this load, but the perfect storm of academically overloaded team members led to my responsibility for nearly 85% of the project by time investment. The other 15 percent was largely devoted by Howard Wang, who was working on training a custom Convolutional network to recognize ripe vs unripe tomatoes through computer vision.