Evolutionary Jumper Simulation

Overview

Solo assignment for the ML2 course at INN. A physics-based evolutionary simulation built with Pygame and Box2D. Each individual agent has a genome of 4 values — jump angle, jump force, body width, and body height — initialized randomly. Each generation, the top 50% are selected as parents, cloned into offspring, and mutated by adding small random values to their genome. Over 80 generations the population visibly improves. A Matplotlib fitness graph plots average and max fitness across all generations showing the learning curve.

My Contributions

• Implemented random genome initialization for 20 agents with uniform distribution
• Built the selection operator — sorts population by fitness and keeps top 50%
• Built the variation operator — clones the best parents and fixed population shrinking bug so offspring always stays at 20
• Implemented mutate() — mutates genome with random noise and clamps values between -1 and 1
• Built the full evolutionary main loop across 80 generations: selection → variation → mutation → evaluation
• Added Matplotlib fitness graph tracking average and max fitness per generation
• Added color mutation inside mutate() so agents visually change color each generation
• Debugged and fixed 10+ bugs: population shrinking, wrong indentation on plt, generation number showing None, duplicate generation 0

Tech Breakdown

Python, Pygame (visualisation), Box2D (physics simulation), Matplotlib (fitness graph), evolutionary algorithm (selection + mutation), genome-based agent design, clamp utility for value bounding