AI Learns to Escape A Wrecking Zone - Deep Reinforcement Learning

Hi Everyone! I posted here a couple of months ago about my plugin UnrealMLAgents and one of my first demo videos.

Since then, I’ve been working on a side project to bring my own character to life, and now I’m excited to introduce Alan — a tripod robot built to take on dynamic challenges, starting with escaping a chaotic wrecking zone. In the video, Alan faces 3 levels of increasing difficulty.

He was trained in Unreal Engine using UnrealMLAgents. Alan uses ray sensors to understand his surroundings, learns through trial and error, and gradually improves his ability to reach the portal while avoiding wrecking balls! The project was inspired by the amazing work from AIWhareouse of course. I aimed to capture a similar feel, but in Unreal with a different character.

Behind the scenes, multiple Alans are training in parallel to speed up learning and avoid hours of watching him get wrecked. !

Would love to hear what you think guys!