h0uter / factor-graph

library implementing barebones gbp for use with factor graphs.

factor-graph

library implementing barebones gbp for use with factor graphs. Based on work

• https://colab.research.google.com/drive/1-nrE95X4UC9FBLR0-cTnsIP_XhA_PZKW
• https://gaussianbp.github.io/

Dev notes

MVP 1

• only a goal layer.
• click to assign a new goal state $x^G$.
• agents can always communicate with each other.
• they optimize to push their goal state a minimum distance apart from the other agents' goal state.

-> done see examples/diverge_from_line.py

MVP 2

• now do it in 2d

mvp 3 create a gui for it.

• make an event loop
• [ ]

Further mvps:

• implement MPC.
• map landscape for mapping discrete action methods.
• make plan for how to operate over a navigation graph instead of over a continuous space.
• go from goal layer to task layer.
• mpc is also just pushing nodes apart that lay too close together.
• information layer.
• so