Abstract
The optimization of pushbacks in open pit mines, which subdivides the ultimate pit into operational volumes, plays a pivotal role in the operational feasibility and economic value of a mining project. Traditionally, pushback optimization relies on generating many feasible pushbacks from which the optimized ones are selected; however, this method depends heavily on manual input and approximate production schedules, limiting to evaluate multiple criteria or adapt to changing conditions. This study introduces an automated pushback selection approach using reinforcement learning (RL) by evaluating operational constraints and economic factors over time. The method maximizes NPV while respecting practical constraints, such as minimum mining width, and is adaptable to include additional considerations. Experimental results using a publicly available dataset demonstrate that the RL approach improves NPV by 6% over the best case strategy. The result shows the potential of RL to enhance automation, robustness, and efficiency in pushback selection.