High-Accuracy Autonomous Excavation Of Free-Form Shapes

Productivity in the construction sector has only grown 1% in the last two decades compared to 2.8% of the world's economy. Additionally, large construction projects run into delays lasting for years more often. In combination with a rising labor shortage, this sector is not performing well even outside of crisis mode. Construction has one of the largest, unused potentials for automation. Exploiting this potential will lead to higher productivity, sustainable construction and new, unseen design possibilities. This thesis aims at automating excavation work, a crucial part of every large construction project. Through automation, the usage of on-site materials will improve sustainability, and the robotic process will allow excavating free-form curved shapes with the same high accuracy as simple straight lines. The soil-bucket interaction forces during excavation vary significantly as they highly depend on the soil composition, density, humidity and many other factors. A successful excavation strategy has to be adaptive to these circumstances. We define a single dig cycle by an end-effector force-torque trajectory. Applying this trajectory to different soil types results in bucket motions used by expert operators to adapt to different soils. In harder soils, the bucket cannot penetrate deeply and it is dragged to fill it. Whereas in softer soils, the deeper penetration allows for curling the bucket to fill it without an extensive dragging motion. Multiple single dig cycles are concatenated to excavate free-form trenches in the Vortex simulation environment. In a second step, we built up HEAP, the world's first autonomous walking excavator. We cover the necessary building blocks, namely sensing, actuation, control and state estimation, to transform an off-the-shelf walking excavator into a versatile, mobile manipulator. Three different actuation variants were realized for the excavator's arm: an electric pilot stage, actuated joysticks and high-performance servo valves. Individual control strategies for the arm, chassis and driving move the machine autonomously with the information from a state estimator for general wheeled-legged robots. The high-performance force control, achieved with the arm servo valves, makes the transfer of the soil independent single dig cycles from simulation to HEAP possible without any changes. In combination with an excavation planner for free-form trenches and an excavation state machine, the same free-form trenches as in the simulation results were excavated with even more accuracy on HEAP. Feedback on the terrain shape is incorporated with an excavation mapping process combining exteroceptive and proprioceptive measurements through a LiDAR and tracing the bucket edge motion, respectively. Finally, HEAP's autonomous excavation capabilities were extended to build free-form curved embankments and combined with the landscape design algorithms of Ilmar Hurkxkens. We created a novel, digitized process for excavation work where survey with a purpose-built drone, design and fabrication are tightly connected and exchange data in real-time. It enables material neutral designs using on-site materials and even allows to account for unforeseen changes during fabrication by adapting the design. An s-curved, 20 meter long embankment was built with high accuracy and smooth surface finish through displacing 30 cubic meters of soil. The material-neutral design was adapted two thirds through the construction process to re-balance cut and fill volumes