Holistic Optimization of Modular Robots

TL;DR

This work introduces a holistic optimization framework for modular robots that jointly optimizes module composition, base placement, and execution trajectory to minimize cycle time for point-to-point tasks. By fusing hierarchical elimination with a unified genetic algorithm that encodes modules, the base as a preface, and IK solutions, the method efficiently searches a vast design space and accounts for feasibility via IK and path planning constraints. Numerical experiments across 300+ benchmarks show that the full scope ($ extbf{m}+ extbf{B}+ extbf{Q}$) consistently outperforms module-only baselines, reducing cycle time by up to $25 eef ext{?}$ and increasing feasibility; real-world validation demonstrates deployment in 9 of 10 cases with limited adaptation. The results substantiate the practicality of jointly optimizing robot architecture, positioning, and motion, offering a scalable route to faster, more reliable modular automation in industrial settings.

Abstract

Modular robots have the potential to revolutionize automation as one can optimize their composition for any given task. However, finding optimal compositions is non-trivial. In addition, different compositions require different base positions and trajectories to fully use the potential of modular robots. We address this problem holistically for the first time by jointly optimizing the composition, base placement, and trajectory, to minimize the cycle time of a given task. Our approach is evaluated on over 300 industrial benchmarks requiring point-to-point movements. Overall, we reduce cycle time by up to 25% and find feasible solutions in twice as many benchmarks compared to optimizing the module composition alone. In the first real-world validation of modular robots optimized for point-to-point movement, we find that the optimized robot is successfully deployed in nine out of ten cases in less than an hour.

Figures (7)

• Figure 1: Overview of this article: The robot task and available modules $\mathcal{R}$ are the inputs to our method, which optimizes modular robots. Our approach jointly optimizes the base of the robot, its module composition, and the trajectory to solve the task. Examples of results from simulations and real world experiments are shown last.
• Figure 2: A simplified example of crossover for a single goal $g$, which allows us to drop the first index from all IK guesses, i.e., within this figure $q_{m_{j}} = q_{ g , m_{j}}$. The different colored boxes show the genes from robots 1 and 2 that are recombined in the new robots 1 and 2.
• Figure 3: All module types inside modrob-gen2 manufactured by RobCo\ref{['fn:robco']} available in our lab. Modules come in two flange sizes: small (left) and big (right). All distal/output flanges point to the lower left of the image.
• Figure 4: Center lines show mean cost (top) and success rate (bottom) with shaded $\qty{95}{\percent}$ confidence interval for the considered test sets (columns) and optimization scopes (color). Statistics are calculated over 23 (real world) or 96 (all other) tasks and five seeds.
• Figure 5: The best robot for the task Around, Seed 4 moving from goal 1 to 2. The 6-DoF robot comprises the modules Base - $2 \times$ D116 - L116-350 - $4 \times$ D86 - I86-350 - Gripper, located at $(x, y) = [13cm, 5cm]$, and rotated by $\theta = 0°$ around $z$. A video is available at\ref{['fn:paper_webpage']}.