Emergence

undergrad thesis on the emergence of multi-agent behaviours

Work Roadmap

Stage 0

• literature review
• reproduce rl, marl, language+rl works
• checkout available environments/simulators

multi-agent works

• MADDPG
• COMA
• DIAL
• CommNet

Stage 1

• easy cooperative, competitve and communication multi-agent behaviors
• learn to interact for more efficient task solving student-teacher framework 2 goals: task solving & being a good teacher

Stage 2

behavior change of protecting eggs is very powerful in terms of evoluation need adversarial agents need to set the constraint just right and conditions complex enough learning constraints automatically? from demonstrations?

Stage 3