Multiple Agents Scheduled Simulation

[sholloway]

Update the existing scheduled simulation to work with multiple agents.

Agent Scenarios

• Agents traversing paths.
  • Looping a Path: Agents moving in concentric circles.
  • Agent moving along a path, turning around and going back.
• Scheduled Attribute Changes
  • Change Color based on some function (e.g. sine wave).
• Agents just spinning around (rotating with no translation). Have multiple agents in a line rotating at different speeds.
• Following: Having multiple agents traverse the same path.

Tasks

• [x] Shared Paths: Have 4 agents on the same path, going the same direction.
• [x] Color: Enable agents to have unique colors.
• [x] Looping Agents: Have four nested closed paths. Have the agents going in opposite directions and speeds.
• [x] Pacing: Have four agents traverse different paths that are NOT closed. When they get to the end of the path they should reverse their direction and retrace their steps.
• [x] Agents could have their orientation follow the path or be focused on another point.
• [X] Rotating: Have 4 agents in a cluster that are all simply rotating around their axis. They should rotate at different speeds.
• [X] Color: Enable scheduling color changes.
• [X] Flashing: Have an agent "flash" their color as a scheduled event.
• [X] Look at using Protocols for handling types and dealing with the renderer function types. I don't like how the simulation context and render layers are defined. Too loose. Passing dictionary objects around sucks. Need to think about patterns like Entity/Component.

[sholloway]

Trying to implement this has led to the place of not knowing how to reconcile scheduled based updates vs a traditional update loop. I've started diving deep into trying to identify the appropriate way to organize the agents.

• Entity-Component Pattern (EC) Game Patterns Book
• Entity Component System Pattern (ECS) GameDev Article

Should the Scheduler be more sophisticated? Look at Naughty Dog's [Job Scheduler Design] Christian Gyrling

• Video
• Slides

Additional Concurrency Options (Fibers)

• Wikipedia
• Greenlets
• python-fibers

[sholloway]

Digging into this has left me with a strong feeling of not being organized. I need to spend more time in diagramming mode before writing any code. Concepts that I need to Flush Out

• What is a frame?
• Frame Parameters data structure (ala Naughty Dog)
• What is an entity in a simulation. Entity-Component/ECS/Something Else?
• Should I deal with Jobs that can spawn other jobs (counters). It feels like just scheduling future running ones is too loose.
• Multi-Frame Paradigm?

[sholloway]

Dale Weiler's write up on implementing a M:N fiber solution is insightful.

In the Python space combining an event loop (e.g. pyuv) with a fiber library like Greenlets or python-fibers may be good enough for my needs.

TODO

• [ ] Try spawning nested jobs using python-fibers/greenlets. How flexible are they? Can native python coroutines be used instead?
• [ ] Are there abstractions for dealing with multiple python processes?
• [ ] Is a dedicated event loop needed? Possible options: asyncio.events.AbstractEventLoop, pyuv.

[sholloway]

Parsers and Generators

Regarding the emergent narrative aspect of this, I wonder if there is a need/opportunity to explore using grammar for crafting the generation of agent behavior, history, or dialogue.

EBNF may be good for defining generation rules.

Some Python tools.

Consider string grammars vs graph grammars.

Graph rewriting

Social Network

Th generation of "stories" can be informed by an agent's social network. Each agent could potentially maintain a list of their relationships. Who they know and what the relationship type is. A relationship between two agents could be multifaceted. Work, friends, lovers, etc.

Being able to traverse and visualize both an agent's network and the network as a whole could be useful.