I’m curious what you think of our approach. Still early but the machine is outside in the fields and even on the shortest day of the year it traveled 20km on solar power.
I think that looks a lot more like the right direction.
I would say that the precision part of Ag robotics is the most important part and the success of this project is likely going to be resting on that entirely.
It boils down to something like:
1. Can you sense and classify (i.e. Deep learning based vision)
2. After that, can you turn a pixel into field coordinates. (Pose-> gps/encoder/imu/etc)
3. Can you act accurately/quick/effectively enough (kill/treat mechanism, electromechnical design)
Then it's can you do it reliably enough. Then it's can you do it fast enough. Then its can you do it cost effectively enough.
Farmers won't mess with anything not reliable. They generally aren't interested in anything with worse performance than existing processes, even if greener/cheaper (unless they are significantly so). They most especially will not mess with anything that doesn't generate value in their eyes and they are very good at understanding what generates value and what doesn't.
Personally I’ve been really pushing for “community supported development”, where a community crowd funds an idea. Like the Godot video game engine or Octoprint. We’re applying to non profit fiscal hosts now. (I expect we may fork the project and have a for profit version in the future, but I’m all about maintaining a complete open source system as a non profit project. See the Dronecode foundation for an example of a split profit/non-profit project.)
I want to have a robot for myself and for others that can manage a one to five acre plot using high volume organic production along the lines of methods popularized by Eliot Coleman, JM Fortier, or Charles Dowding. I’m leaning towards no-dig (Dowding) but it will take a lot of experimentation to see what really works best.
But I say all this because I think a lot of companies are going to be focused on meeting market demand for existing farming needs. I’m hoping that for our non profit work we are pushing the envelope for highly ecological development, supported by enthusiastic believers in our mission. I’ve been learning about reduced till and no till agriculture and it seems super promising for a robot like ours. My hope is that long term, we can help build a system that allows these techniques to scale.
That said, the need for a precision vision system is even greater in this scenario. There’s going to be a lot of variety in what our system needs to identify and do! Luckily vision is the thing I really enjoy learning. I think if I can drive prototyping and I can get a broader community going, we have a good shot of getting a whole system going. At the very least I think our open source vehicle is very good, and will be a solid base platform for people once we release the V2 research vehicle.
If you have that much excess power I wonder if you could solve the precision water problem with tanks that top up by wringing it out of atmosphere in humid climates.
I would say that the precision part of Ag robotics is the most important part and the success of this project is likely going to be resting on that entirely.
It boils down to something like:
1. Can you sense and classify (i.e. Deep learning based vision)
2. After that, can you turn a pixel into field coordinates. (Pose-> gps/encoder/imu/etc)
3. Can you act accurately/quick/effectively enough (kill/treat mechanism, electromechnical design)
Then it's can you do it reliably enough. Then it's can you do it fast enough. Then its can you do it cost effectively enough.
Farmers won't mess with anything not reliable. They generally aren't interested in anything with worse performance than existing processes, even if greener/cheaper (unless they are significantly so). They most especially will not mess with anything that doesn't generate value in their eyes and they are very good at understanding what generates value and what doesn't.