Laser Guided Assembly — gridguidance

GUIDANCE

I practice architecture by
coding constructive processes

Geometry Engines
Feedback Loops
Automation

Laser Guided Assembly

Cambridge, MA
MIT, Master of Science in Architecture Studies Thesis

Advisor: Skylar Tibbits
Reader: Caitlin Mueller

Does digital optimization have to stop when construction begins?

The standard paradigm for optimizing a prefabricated truss is as follows: the performance and placement of inidivdual pieces are considered inside a computer simulation, and the design is projected onto a work surface for builders to locate the pieces. Once the information leaves the digital model, the optimization stops.
The Top-Down Approach
After disconnecting from the digital model, it is the responsibility of the builder to make small adjustments, tuning tolerances and keeping dimensions from drifting around a structure. The digital model and the physical construction drift apart, with no mechanism in place to bring them back together.

A Bottom-Up Alternative
An optimization I would prefer would be for the moves to be given one at a time, allowing the assembly process to be guided along a desirable route.

How do you represent architecture as a series of moves?

Move Notation
Designing what constitutes a “move” In order to convey a design as a series of moves, we need to choose what constitutes a “move.” The implementation involves building a custom class structure to keep track of trusses, emphasizing connection between nodes rather than particular coordinates

Move: Place_two
Implementing a framework for storing designs as “moves” rather than as pieces

Long Exposure : Previewing the next 3 moves
What would it look like to suggest the next move(s) to a builder?
What if we could convey a design as a series of moves for the builder to make? By serving the instructions incrementally, we open the possibility for the builder’s corrections to be incorporated into a continuous digital optimization.

Building a Move-Order Solver

The Digital Circularity Toolkit (DCT) Hungarian Matching algorithm is one approach to find the best fit for each bar of the graph, matching each bar of the graph with an inventory of available parts. In practice, the best match is never precisely the right length: the problem is overconstrained.

Consider that if we really don’t want to have to modify every single bar, we might prefer to find the best order instead. The truss can be initialized with certain locked nodes.

Consider that the order of placement affects the subsequent placement.
Then applying Dijkstra’s algorithm, we can solve for optimal move order between locked nodes

Trade-Offs of the Decision to Lock Nodes
For a given truss topology plugged into the move-order solver, more constraints can be added to keep the dimensions from drifting away from the intended design.

If the designer allows the builder more freedom to pick the best moves, fewer modifications are required.

Move Order Solving a 50-bar Cantilevered Truss with zero modifications
As proof-of-concept to determine the feasibility of the move-order-solver, we apply the solver to a Michell Structure*

* Michell structures as theorized by Anthony Michell are optimal structures representing a continuum of orthogonal stress trajectories, commonly referred to as lines of principal stress. The simplest design of the truss would require only two bars, yet the theoretical optimum could be approximated with an increasing number of bars to converge toward the ideal solution: a truss with infinite bars to perfectly carry a given load.

Communication Schematic: Live Two-way Digital Twin
Because the optimal “move-order” is dependent upon prior construction, the architect cannot explain the design with a static paper drawing. The means of communicating design intent needs to be responsive to updates on the construction site, and the design model needs to update in response to changing conditions.

Hardware Development: Laser Projector
In order to send digital instructions to the construction site, the architect needs a physical endpoint stationed on the site. We imagine that the laser level found on any ordinary construction site can become the destination for the architect’s instructions.

Basic functionality: X-motion
A stepper motor controls the rotation of a mirror, reflecting a laser beam.

Controlling the rotation of two stepper motors is enough to create XY motion, sufficient to draw shapes in a 2D plane.

Projection Pipeline
The move-order optimization occurs in a true-to-scale design plane. In order to draw the shapes correctly, the projection needs to correct for the angle of the laser projector relative to the ground.

Laser Projector Calibration
Testing to ensure that the laser can return to earlier locations.

Homing Routine
Scaling the projected design with the physical design space to ensure that projected moves will match the reality of the construction site.

Verification Routine
The laser projector outlines the placement of the first two bars, confirming the validity of the first move.

Field Correction
When the builder diverges from the design, they can mark the location of the new node by “scanning” it with the laser. Pointing the laser at the new location, the motor positions can be read and backwards calculated to update the design change in the digital model.

Timelapse: Laser Guided Construction of a 512 bar Michell Structure