High Precision Excavator Control

追加: 2026-05-17

[00:00:03]In this work, we present [music] an autonomous control stack for heavy-duty grading with centimeter accuracy.

[00:00:09]State-of-the-art semi-automatic grading controllers fail under heavy load by losing precision and not using all torque available to the excavator.

[00:00:21]Different hydraulic architectures make it challenging to remove substantial material while maintaining a high precision surface.

[00:00:28]We demonstrate our approach transferring between two common hydraulic architectures, load sensing and negative flow control, with the same in-soil accuracy. For adaptability, we split the control stack into two parts.

[00:00:40]The hydraulics-aware joint velocity controller adapts to the different architectures and provides accurate tracking of the target joint rates.

[00:00:48]Second, we use a model predictive controller for path tracking. It coordinates the joints to follow the design surface. We validate the system on two full-scale excavators, a Menzi Muck M445 for load sensing and a Case 250 for negative flow control.

[00:01:09]We command the velocity to the joint low-level controller under different joint loading conditions. The accurate tracking with different resulting valve commands demonstrates our method's ability to compensate for the pressure-dependent joint dynamics found in negative flow control machines.

[00:01:24]Grading experiments show that our controller tracks the design surface with an accuracy of about 1.8 cm independent of the cutting depth.

[00:01:33]Stalling occurs only at the maximum function pressure.

[00:01:36]Repeated experiments on the load sensing machine confirm the accuracy and stalling behavior.