Open ghost opened 5 years ago
Should do some research on stage, stdr, and flatland to see if we can (or even should) leverage existing lightweight 2D ROS-compatible simulators. I'm not very familiar with this issue, but I suspect we would have to see if any of them offer the balance of control over everything that we need while also providing useful tools for us to use.
No idea if they'll be useful. But it's worth looking at existing packages.
The simulators seem great right now, but as far as I can tell, they would be too much for our highest-level simulation (the MDP) and so I don't know if they're what we want. But perhaps we could leverage like you mention. I'll take a deep dive into a few of them that have the most stars and most active development.
📚 Meta Issue
We need to create a visualization platform for our initial 2D network training. This will allow us to watch a match play out in 2D. This issue DOES NOT cover the scope of tournament play, and a separate issue will be created once tournament play is implemented in the training software. Since all of our networks will, as far as we can predict at the moment, begin as simple 2D networks, this platform must be sufficiently abstract to allow easy implementation of future games.
Tracking
33: Create a blank rqt window.
34: Find and render the most accurate top-down 2D representation of the map.
35: Implement and render a match timer, with start/stop and fast-forward/rewind functions.
36: Render all possible states (positions w/ actions) and only transitions from the current state.
37: Render the current robot position (at a state/along a transition).
38: Render the current action if at a state.
39: Render the positions and state of current hatches/cargo.
40: Render the score view (copy the FMS view from the manual when it is released)
48: Connect agent node and visualization platform.