Repository detailing ongoing issues and challenges facing the system.

Table of Contents:

Nozzle alignment

The 32 nozzles in the printhead are not perfectly straight. This means that dispersed fluid does not always go directly straight down, which is particularly notable at greater printhead heights as it increases the deviation. This is currently attempted to be addressed through software calibration to adjust the expected and actual droplet position for a given height. In future systems, it may be possible to request systems with narrower tolerancing.

This is not a manufacturing defect, as the printhead was designed to print at <1” heights where this issues is not noticeable. It is not possible to straighten the valves in existing systems. Response from REA Jet when discussing this issue on 14 June 2023. “The nozzles are jewels that are pressed into place with the tolerance of printing no more than an inch away from the substrate to be printed on. Unfortunately, the nozzles are not adjustable. The heads are made for printing large text on industrial material, so the precision outside of printing in the intended manor was not thought about.”

Velocity estimation

For the system to accurately dose weeds it needs to have a high precision estimate of the velocity of the vehicle - to predict when the valve will be over a target plant given the forward facing image. The system is highly sensitive to errors in velocity. For example, at nominal velocity of 1 m/s, a 1% error in velocity will result in a ~10 mm longitudinal error in spray position.

In the current prototype (as of writing 2023-10-10) the vehicle speed is determined from the Amiga control and then manually calibrated in the software to improve accuracy. In a platform agnostic system it would be ideal to have local measure of velocity that does not rely on an external system on the vehicle. Estimating velocity from images (as is done by Verdant) would be the preferred option as the system will always have a camera onboard - this is being investigated with the require computational time being the greatest concern. A gauge wheel with a speedometer (e.g. using a rotary encoder) would be another option. RTK GPS initially appears less feasible in terms of estimate error.

In all of these systems, if the vehicle can reliable travel at a precise and consistent speed, then it removes some of this continuous velocity estimation load as the system can just be calibrated to that specific velocity of operation.

Mechanical alignment

The targeting system relies on the printhead, camera and and the vehicle to be aligned and square. In the current prototype the mechanical attachments were designed to allow some mobility for calibration and testing. However, in future systems methods ensuring the mechanical mounts are always aligned would be useful.