A Reusable Simulation Pipeline for Many-Agent Reinforcement Learning
Date:
I presented our paper “A Reusable Simulation Pipeline for Many-Agent Reinforcement Learning” at DS-RT 2024.
The talk described a reusable experimental pipeline for studying reinforcement learning in systems composed of many interacting agents. It highlighted the role of simulation in separating environment modelling, learning logic, experiment execution, and result analysis, with a prototype built on the Alchemist simulator.
The presentation focused on reproducibility and scalability, showing how a structured simulation workflow can support the systematic evaluation of learning strategies in large distributed systems.