Can ML-Agents Train a Real Robot? A Unity Sim2Real Experiment
In this sim-to-real experiment, I use ML-Agents to train an AI in a virtual environment, then deploy it to a physical turtle bot. The AI successfully steers the real robot to randomly placed 'targets' - an example of how game engines can be used as tools for robotics development.
💬 Join our Discord community (free!): / discord
▬ Support My Work ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
❤️ Support me on Patreon: / ludicworlds
☕ Buy me a coffee: https://ko-fi.com/ludicworlds
Thank you for your support!
▬ Timestamps ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
0:00 - Intro
0:36 - Project Overview
1:55 - Building Real Bot
2:35 - Gamepad Control of Bot
2:51 - Training Simulated Bot
4:16 - Tracking the Real Bot
4:51 - AI Agent Controls Real Bot
5:17 - Moment of Truth
5:52 - Outro
▬ Useful Links ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
► "Get Started with ML-Agents in Unity" Playlist: • Get Started with ML-Agents in Unity
► Keijiro's 'AprilTag' package for Unity: https://github.com/keijiro/jp.keijiro...
▬ Credits ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
► Music by - CO.AG Music: / @co.agmusic
#Unity #mlagents #Sim2Real #reinforcementlearning