Autonomous Agent with EWC (Continual Learning)

This repo implements a simple REINFORCE policy-gradient agent that trains sequentially on multiple OpenAI Gym tasks while using Elastic Weight Consolidation (EWC) to reduce catastrophic forgetting.

Quickstart

1. Create virtualenv and install:

This script trains on CartPole-v1, computes EWC constraints, then continues training on MountainCar-v0 using EWC.

Files

• main.py: Orchestrator / pipeline
• train.py: Training loop helper
• agent.py: Agent implementation (REINFORCE)
• model.py: Policy neural network
• ewc.py: EWC implementation (Fisher estimation + penalty)
• utils.py: env compatibility helpers, plotting, saving