19th
Discussed topics on slides:
Link to slides (Rafael)
Link to slides (Angelo)
- General remarks
- Update on To Do
- Brief discussion on dynamic agent creation (see issue #1)
- Brief discussion on the challenges in rounds with different number of agents (15x30x50)
- Reactive vs Proactive behaviour
- Angelo's slides on 2019 agent identification
- New changes to agent identification
- Resource hungry, can we limit this somehow?
- Discuss A10 in picture below
- how many tasks can an agent accept
- An agent can only accept one task at a time, subsequent accept actions will replace the task accepted (we will need an internal belief to keep track of this)
- team B starting position
- Mirrored starting position (cluster size based on the clusterSize parameter on sim.json)
- how much do tasks decrease over time
- Rewards always decrease by a percentage up to a certain threshold (also percent based)
- Probability of agent identified? First measure max step all agents identified
- Add goal positions to the identification process
- Switch to proactive model, using Moise to coordinate each phase
- Think about how to deal with circular grids (we can actually calculate the size by leaving an agent from the initial cluster in the same position it started and sending others to opposite directions, when they meet the same agent again we know the size of the map)
- Do we need a global map? Try to calculate new information based on the position the agent was first identified in another map