Deep Reinforcement Learning in Mining Fleet Optimization

Reinforcement Learning (RL) can drastically improve fleet assignment decisions in mining. In my project Vivania, a DRL agent learned to reduce queue times and exploit system dynamics to increase throughput with only CPU training.

Key takeaways

• Reward shaping should balance travel time, queue avoidance, and throughput.
• Sim fidelity matters: include realistic constraints to avoid degenerate tricks.
• Evaluate with domain KPIs: tons/hour, queue length, idle time, fuel burn.

Explore the repo for a starting point to experiment with DRL in mining operations.