Robot excavators learn to move soil like human operators
Size Doesn't Matter: Material-State Reinforcement Learning for Excavator Transferable Soil Manipulation
Robotics
Summary
Moving dirt with machines like excavators is hard because soil changes shape and isn’t like solid objects. The authors trained robots using a physics-based simulation so the machines can handle soil more like people do, using all parts of the shovel and adjusting for soil shape and compactness. Their program works on big machines and small robots without retraining, and it successfully built a large embankment automatically, matching human experts in speed and quality. This shows robots can learn to work with soil more flexibly and reliably.
What this means in practice
- •For construction automation teams: Automate earthmoving tasks on large excavation machinery using learned policies that handle varied soil states for reliable embankment building.
- •For robotics engineers: Develop robotic soil handling tools that transfer learned behaviors from simulation to varied robots without retraining by normalizing actions and calibrating interfaces.
Authors
Lennart Werner, Pol Eyschen, Sean Costello, Pierluigi Micarelli, Andrei Cramariuc, Marco Hutter
Abstract
Earthmoving tasks such as excavation, backfilling, or embankment construction require deliberate repositioning of deformable soil. For these tasks, human operators use all shovel faces, while autonomous systems so far are limited to excavation and dumping. Current methods often rely on heuristic models but do not incorporate soil mechanics. We address this shortcoming by using Reinforcement Learning in a GPU-parallelized Material Point Method particle simulation. Our controllers are conditioned on material state such as shape and compactness, enabling skills that use multiple contact faces of the tool and displace material both inside and outside of the shovel. To use the same learned weights across machines, our policies operate in a normalized end-effector space and are deployed through a calibrated machine interface. We evaluate this calibrated transfer on an 11.5t hydraulic excavator and a 500g tabletop robot. We validate performance through autonomous construction of a 42m long, 2.1m high embankment in 45min, executing 201 individual policy strokes without failure, retry, or operator intervention. In a direct comparison, the autonomous controller matches an expert operator's progression speed and produces a higher, more consistent embankment. Additional qualitative backfilling and compaction experiments demonstrate the material-state awareness and calibrated transfer across machines.