Autonomous Dry Stone: Mobile Robotic Construction with Naturally Nonstandard Materials

Abstract

On-site robotic construction not only has the potential to enable architectural assemblies that exceed the size and complexity practical with laboratory-based prefabrication methods, but also offers the opportunity to leverage context-specific, locally-sourced materials that are inexpensive, abundant, and low in embodied energy. Toward these ends, this doctoral research is focused on developing a novel process for the robotic construction of dry stone walls in situ, bounded by design constraints and facilitated by a customized autonomous hydraulic excavator. These walls are built using as-found natural stones and reclaimed demolition debris, using a construction pipeline that automatically collects an inventory of these materials by detecting, grasping, and 3D-scanning them directly on site. Given a limited inventory of these digitized stones, a geometric planning algorithm determines how each of these objects should be positioned toward the formation of stable and explicitly-shaped structures. By adapting knowledge from traditional stone masonry practices, this planning algorithm uses a combination of geometric features to seed hypothesis stone placement candidates. These candidates are then refined toward stable and geometrically-aligned solutions using a combination of torque- and penetration- constrained iterative closest point registration, and physics simulation. Ultimately, these solutions are classified for placement viability, using a supervised model that considers a 3-channel signed-distance-field data-representation of each solution that encapsulates the candidate stone, the local context of terrain and previously-placed stones, and the freeform target-wall geometry. To accommodate settling and process tolerances, the geometric planner works iteratively, using information from an accumulated LiDAR map to regularly update the as-built structure after each stone is placed—and before each successive search for new candidate placements. Using this approach, the planner is able to inform the construction of double-layer walls, using highly nonstandard stones and debris—creating structures with a 60% fill-to-void ratio within arbitrarily-defined wall boundaries. This process is further informed by large-scale, outdoor physical experiments. These experiments resulted in the construction of three robotically-constructed dry stone walls, that are built using gneiss boulders, erratics unearthed on construction sites, and salvaged concrete demolition debris. These demonstrators include a 40-stone s-curved wall (5 x 1.6 x 3 m), and a linear freestanding wall (10 x 1.7 x 4 m) constructed with 24% reclaimed concrete. At the last stage of development, this work is evaluated through the first robotic construction of a permanent and publicly-accessible stone retaining wall (65.5 x 1.8 x 6 m) consisting of 938 unique elements—and that is integrated with robotic landscape features based on the doctoral research of Dominic Jud (Robotic Systems Lab) and Ilmar Hurkxkens (Chair of Landscape Architecture). Collectively, these demonstrations saw the robotic placement of over one thousand coarse boulders, with each weighing an average of one tonne. The physical testing conducted during these experiments revealed shortcomings and necessary improvements to the process, and allowed us to provide the first benchmarks for large-scale robotic assembly with nonstandard stones. These studies demonstrated robotic stone placement rates up to 12.2 min/stone, and quantified the ability of this method to reduce emissions by upwards of 40% when compared to equivalently performing concrete structures. This work illustrates the potential of autonomous heavy construction vehicles to build adaptively with highly irregular, abundant and sustainable materials that require little to no transportation and preprocessing—creating structures which benefit aesthetically and environmentally from the properties of regionally-specific natural materials.