Background

In this page we briefly describe what the BenchBot is, its beginnings and why we decided to embark in this project.

What is BenchBot?

BenchBot is a high-throughput robotic platform which can collect data in a systematic, fast and reliable manner and can be adjusted to greenhouse and semi-field settings. The system consists of an autonomous platform to which different sensors can be mounted, an imaging system (camera + lighting), a user interface and a data processing pipeline, all of which we explain in depth in this document.

This is an open source hardware and software project that we hope other people in the ag and tech community get to use as is or that can be used as a starting point for developing other tools that adjust better to other setups.

How everything started

In the year 2019 we were running a breeding project in one of the greenhouses at BARC which required measuring the height of 960 plants growing in pots weekly. This task required 2 people doing this activity for a full day. We were also collecting overhead images of the plants using a PVC frame and a commercial small digital camera which also took a full day and two people to accomplish every week.

Crimson clover shade tolerance experiment at one of the greenhouses at BARC, MD.

Example images of crimson clover at (1) week 8, (2) week 11, (3) week 14 and (4) week 17.

The time that this took and the quality of the images that we were getting made us realize that we needed a better system. And that is why the BB was created.

BenchBot v1

The first version of the BB was design to collect images in a semi-automated way since the plate to which we mounted the camera moved on its own, but we had to manually move the platform from one row of pots to the next. This drastically improved the efficiency of data collection and the quality of the images that we were getting for the second year of the experiment (2020). We were able to cut the data collection time in half and we were getting images of individual pots instead of several pots per image.

Example images from crimson clover at week 6 (1), week 8 (2), week 10 (3) and week 12 (4).

For this project we used the intel RealSense d-435i camera which collects RGB and depth data from which biomass could be estimated.

BenchBot v2

The next milestone was to fully automate the BB so the system could become more efficient, reliable and require less labor. As we were planning for these BB improvements, a very ambitious project was starting to brew within our research group, a Weeds Image Repository for the United States for which the BB has now become a fundamental tool.

This new project required the BB to be able to perform in a semi-field setup so we started working on a design that could accommodate that. By the spring of 2021 we had a fully automated system. With the addition of motorized wheels, an object avoidance system and a custom UI, the BB was able to move fully on its own from pot to pot and from row to row while keeping track of the collected data. We were able to bring the data collection time down to a fraction of what it took with the BB v1.

With the semi-field setup that you see in the video, BB was able to collect 180 high resolution images/hour, taking ~2.5 h to scan this full setup of 450 pots.

BenchBot v3

Coming in 2023 2024

For this new version of BB, we are looking to increase throughput, reliability, and ease of use.

Full hardware and software redesign using more robust and available parts.
Large, configurable gantry to eliminant needing to turn around.
Based on Farm-ng Amiga for Y-axis traversal. Eliminate the need for tracks & provide a robust platform to build on.
New industrial style camera that is easier to get data from and survives the elements better.