Roadmap

[thatprogrammer1]

Compiling a list of things that probably should be worked on to get a decent full-stack system working

Environment

• [ ] Add colors to prisms
• [ ] Get prisms working
• [x] Fix prisms sometimes falling out of the camera sight at init

Perception

• [x] Figure out a non-cheaty way to do the segmentation #7

Grasping

• [ ] Grasp in the center of blocks, rather than the edges
  • I think this relies on #7, but maybe there's a way that I'm not thinking of?

Placing

• [x] Place so COM of prism is at the center of the stacking cylinder
  • Estimating COM can be done with #7, or possibly it can be done by measuring the torques on the joints as Tedrake mentioned?
• [ ] Place on the stack softly, instead of pushing down on the stack
  • I can think of 2 approaches:
  • 1. Estimate the bottom of the block, and just plan so that the bottom of the block perfectly aligns with the top of the stack.
  • 2. Force control something something - descend until you feel the block touch the stack

Planning

• [ ] Figure out what to do when blocks spawn in the stacking cylinder
  • Perhaps just choose a stacking cylinder at start that doesn't intersect with other objects? Or we can have a clearing state for any blocks that are not perfectly in the cylinder?

[thatprogrammer1]

Things that need to get done:

• [ ] placing on stack softly @thatprogrammer1
• [ ] Clearing out stacking cylinder (being worked on by @kehemo in #11 )
• [ ] measuring performance - I imagine this looks like running X numbers of offline simulations and having success criterion @thatprogrammer1

Lots more would be cool and doesn't feel like we're that far but maybe taking classes that end up with 4-5 projects due in the last week was a bad idea :cry:

Here's some cool stuff if we're feeling spicy (but i don't think is really necessary for finishing out the project):

• [ ] making more interesting stacks than being tall (making leaning stacks?) I actually don't feel like this would be that hard due to kenneth's COM sensing
• [ ] prisms uhhh lol (at this point let's just hand create a few prisms? and just put them in) @AnirudhRahul is gonna work on
• [ ] this goes along with prisms kinda but our bot isn't really able to detect what side is which for rectangular prisms. it needs to be able to grasp on the sides that aren't parallel and then turn it so that the paralel sides are on the ground. This is also necessary if we want to try to be able to get tallest stacks possible for the rectangular prisms.