Full Paper

Supplementary Video

The structure of the proposed framework from optimization of a robot model to mission execution. The bold boxes indicate the contributions of this work.

In simple terms, the automation process works as follows (we highlight points of human intervention):

• A BIM model provided by an operator is parsed, and necessary information for simulation is automatically extracted using the ROSBIM framework.

• The operator can use a handheld device with an intuitive interface to generate a mission; for instance, they specify drill poses or patches of a wall to be painted.

• The generated mission is automatically translated to a formalized XML task description, following the standardized schemata introduced in CoBRA.

• This triggers the execution of a multiobjective lexicographic algorithm (Robot Synthesis) for robot morphology optimization. Simply said, this algorithm determines a valid and performant assembly of available robot modules to execute the mission specified earlier. It considers not only performance in simulation but also other objectives, such as robustness against calibration errors and the time needed to perform the physical module assembly. Because these criteria do not allow for a natural hierarchy, we identify a Pareto front of optimized solutions and provide all of them (usually between 2-12) to the user.

• The operator chooses a Pareto-optimal solution and physically assembles the robot. This takes between 5 and 20 minutes, depending on the chosen configuration and the previously assembled robot.

• The mission is dispatched and automatically executed. The robot navigates the construction site with the help of the provided BIM model, calibrates itself using visual markers, and performs the motion plan that was computed, based on the mission. Thereby, the plan is adapted online to account for possible placement errors.