PredPreyGrass
This page describes the conceptual mapping. For the actual trial-by-trial results — what has been tried, what worked, and what remains open — see the Darwin/Baldwin Trial Log.
Energy and Entropy in PredPreyGrass
Energy flows through the simulated ecosystem from grass to prey to predator, and — as in any real physical system — that flow is one-directional. Energy is never recycled, only degraded, in line with the second law of thermodynamics. This page explains how that plays out in the current implementation of the base environment.
What Counts as Evolution in PredPreyGrass?
Populations can change over time for more than one reason, and it matters which one is doing the work. This page draws a precise line between three kinds of change that can look similar from the outside — ecological population dynamics, within-lifetime learning, and true Darwinian selection — and uses it to say exactly what each PredPreyGrass variant implements.
Challenges of Multi-Agent Reinforcement Learning in PredPreyGrass
Multi-agent reinforcement learning is a promising tool for studying emergent behavior, but it is genuinely harder than single-agent RL, not just a bigger version of it. Building PredPreyGrass surfaces most of the field's core difficulties in a concrete form.