Soil-Adaptive Excavation Using Reinforcement Learning

In this letter, we present an excavation controller for a full-sized hydraulic excavator that can adapt online to different soil characteristics. Soil properties are hard predict and vary even within one scoop, which requires the encountered conditions. The objective is fill bucket with material while respecting machine limitations prevent stalling or lifting of machine. To end, train control policy in simulation using Reinforcement Learning (RL). interactions modeled based on Fundamental Equation Earth-Moving (FEE) heavily randomized parameters expose agent wide range learns output joint velocity commands, be directly applied standard proportional valves real We test 12-ton types soils. experiments demonstrate changing conditions without explicit knowledge parameters, solely from proprioceptive observations, easily measurable.