I have several pockets I need to mow on my property. Often these are dead-ended and are hemmed in by shrubs, fences, the pool, and so on, where if you mapped it as an Area with a border tucked in there or a big Area with No-go Zones, the rover would be unable to turn or get close to the edges of the pocket to do a more complete mowing job.
The workaround is to map an Area in that pocket with a “best effort” – if you can – and add Deadends for anything the rover may have missed. Sometimes you can’t map a useful Area inside a pocket and have to create several Deadends to mow it.
A Deadend Area allows you to map the pocket you wish to mow in a from-to direction so there are no turns and there is better coverage edge to edge, end to end.
A Deadend Area is mapped by creating Deadends on either side of the pocket as deep as needed. The App creates a Deadend Area based on the start and end points of the two edge Deadends, creating the necessary Deadends in parallel between the edge Deadends to mow the pocket.
When mowing, the rover processes each Deadend in turn, mowing the pocket about as completely as it can do.
Examples of candidate pockets:
Pocket next to the “Yarbo Hits Me Too Often” Rubbermaid bin:
Sample implementation. Yarbo is shown trying to mow this as an Area. It does kinda OK but leaves spots in the fence corner and at the back fence. Deadends would have the mower go straight toward the back fence and get most of everything.
We’ve requested more control over the lines the snowblower takes for sure. I have one Area that needs to do a Deadend Area type thing, blowing snow in one direction toward one end. Now that we have zero-turn Sidewalks, there ya go.
No reason Deadend Areas couldn’t be used across modules.
And you know those posts on FB looking for how to mow a trail? And my ask for Double (and more) Width Pathways? Here ya go. Just a small feature? Nope. Obviously the code would have to work out the geometry when the edges aren’t parallel, but that’s just math, and they already do it when mowing Areas with irregular borders.
I would love for Yarbo to just recognize areas like this and mow them dead-end style, because that’s how it should be mowed. If Yarbo did implement a feature like this I would use it, but an ideal implementation would be for Yarbo to take an area and figure out the appropriate way to mow all features in that area. Corners, corridors, weird shapes, whatever. I don’t want to have to figure out an optimal configuration, when an intelligent robot is doing the work for me.
Yarbo, please do better with the software, and make this thing as smart as you advertise. It should “just work”
Having done the Software Development thing throughout the years, yes, a “be smart and just do it!” is the ideal end goal, but I’ve been conditioned to think in “baby steps,” or your only release as a company will be the employees when the cash runs out.
A Deadend Area is likely an easy feature based on existing tech, a human bootstrapping it, and some math (although nothing ever turns out to be ‘easy’ in Software Development).
A Lawn Discovery feature to achieve automatic mapping will need to be put on the Tesla self-driving level, and you’re probably not going to have that much video to train it nor processing power to run it. Same with Vision, which is struggling a bit but moving forward, and there are big enough expectations with just that feature.
@Ken your response got me wondering what the computing power is on Yarbo. I assumed the obstacle avoidance system uses ML. Route planning could be ML, but it’s possibly not. Looking on their marketing website I can’t seem to find any specs.
If Yarbo does have the computing power to run ML and take advantage of all those sensors, then it’s just a matter of time and development. Simulation has allowed dev teams to drastically reduce how long it takes to train a model, not requiring 10M+ samples like it used to.
If Yarbo does not have that computational power, and everything is algorithmic, then this version of Yarbo will be limited for sure.
Yarbo’s ambition does seem to be more in line with the “it just works” mindset. At least that is how they are advertising it. If they gave us more knobs and switches initially, I’ll take it, but I do hope they aim for autonomy over configurablity in the long run.
I have been able to overload something (processor? communications? BT?) by some twisty driving pulling a load up a muddy incline with the controller and having the Bluetooth speaker playing. So there are limitations, which is why I threw it out there as a potential issue if doing AI on-board.
I second this feature request. Right now I have to program 50 parallel lines using Deadends to mow a hill. This would be a huge time saver!!
Picture is attached that show the painted lines…
@Ken, seems overly complex - map the critical 2 steps fence and pool - then a simple Dupe Left or Right until the map shows the area is covered, but whatever the solution, creating half a dozen or as @Offgrid faces, 50!!! dead ends, is painful especially when it’s tough to drive a straight line. Looking forward to that “Drive Straight” implementation as well.
Signed “Wigglepath”
No, not “dupe L/R” – the feature should use math and smarts and stuff to fill in the Deadend Area. Instead of going straight in by the pool on the right, I’d like to follow the contour. Then way over on the left it’s straight in for the fence. Yarbo should compute the Deadends to cover that pocket with what it needs from the contoured side to the straight side, using the Deadend out-and-reverse method.
The secret use case for this is N-wide Deadends to mow trails, and I’d like to get Zero Turn Sidewalks to do the same for the snowblower. These aren’t necessarily straight. This feature will also help with slopes. So it’s a feature with some hefty use cases that add up to quite a bit of functionality.
Agreed. Dupe sounds complex vs just mimic the slope area like the snow blower would be a better option. I believe something like this is in the works or at least being considered.
Maybe I’m completely wrong, but I thought a dupe would be extremely easy as interim assist. If this thing is moving around based on GPS, then in my own simplistic mind I figured a dupe would simply be a certain lat and long plus or minus the width of the rover in deg and sec ??? I don’t mind mapping the dead ends once. It’s a pain yes, but not that bad for me personally. However if in the fullness of time you’re having to redo them consistently after updates or issues, then it could be very painful. I thought dupe would be a relatively easy approach but maybe not. Just sharing ideas.
It might be, but I have a feeling it’s probably not as simple as it sounds. In the end, these types of things need exact precision and arbitrarily dropping one might put it in a situation that the user didn’t intend. I can see the utility of this, but the risk might outweigh the gain. Especially for a new user who doesn’t yet fully understand the system and its nuances.