Evolution

A project for me to explore how Artificial life work in a predefined system and also be able to check out what does evolution mean

so what i have achiveed in here is that i am using pychunk to stimulate the body its just a dog at the moment with 4 legs u already know with reward and finess scores since thats what its job is

the brain part is just a small Neutral Network with very less paramters like 256 and then its algorithm currently i just the best ones right now it takes 5 best ones and then crossovers 30 % so that it can take a new and differnet steps and mutation is 70%

if u wanna run it do python main.py --render it will run the best ones and yeah no AI slpp btw this si what the project is about am also adding a video or gif to show you 8/26/2026

🗺️ Project Roadmap

• ✅ v1 — Straight-Line Walking (Achieved): 2D Pymunk quadruped physics, CPG rhythm inputs, fall termination penalty, and idle reward fix to prevent reward hacking.
• ✅ v2 — Multi-Core Optimization & Web Deployment (Achieved):
  • Multi-Core Training: Parallel rollout evaluation with Python multiprocessing (--workers N), accelerating training by 4x-8x.
  • Live Web Dashboard: Real-time 60 FPS HTML5 Canvas creature animation streaming over Server-Sent Events (SSE).
  • Interactive Controls: Start/pause evolution, tune mutation rates and population sizes dynamically from the web interface.
  • Zero-Config Docker: One-command deployment via docker compose up with volume persistence.
• ⏳ v3 — Procedural Terrain & Obstacles: Randomly generated hills, slopes, curves, gap jumps, and spike hazards.
• ⏳ v4 — Horse Morphology & Rider: Evolving body structure into a horse with a balanced human rider on top.
• ⏳ v5 — Weaponry & Vision: Equipping the human rider with a spear and directional vision sensors.
• ⏳ v6 — Multi-Agent Battle Arena: Multi-creature tournament battles with combat physics and competitive evolution.

🚀 Quick Start (v2)

1. Interactive Web Dashboard

python app.py

Open http://localhost:8000 in your browser to view the live physics canvas, live convergence curves, and start/pause evolutionary training runs with one click!

2. Multi-Core CLI Training

# Train for 100 generations across 4 parallel CPU workers
python main.py --train --gens 100 --pop 50 --workers 4

# Render the best evolved model in PyGame
python main.py --render

3. Analytics & Progress Visualization

# Plot latest training run metrics (generates PNG & HTML dashboard)
python scripts/plot_logs.py

# Live monitor an active training run
python scripts/plot_logs.py --live

4. Docker Deployment

docker compose up --build