Design Optimizer for Soft Growing Robot Manipulators in Three-Dimensional Environments

TL;DR

This work advances design optimization of soft growing robot manipulators into three dimensions by formulating a multi-objective problem over design and configuration parameters and solving it with Evolutionary Computation enhanced by Rank Partitioning. The approach yields a single, manufacturable design $\,\delta\,$ that can realize multiple target-reaching configurations via inverse kinematics while respecting kinematic, manufacturing, and obstacle constraints. Key contributions include (i) a 3D mathematical formulation based on Do et al.'s stiffness-controllable continuum links, (ii) a dynamic genotype-to-phenotype mapping that optimizes link usage per target, and (iii) a robustness comparison of GA, PSO, DE, and BBBC under Rank Partitioning. Empirical results show GA delivers the most reliable, high-precision designs across three increasingly complex tasks, demonstrating the method's practical potential for pre-manufacture design of safe, adaptable vine-like robots.

Abstract

Soft growing robots are novel devices that mimic plant-like growth for navigation in cluttered or dangerous environments. Their ability to adapt to surroundings, combined with advancements in actuation and manufacturing technologies, allows them to perform specialized manipulation tasks. This work presents an approach for design optimization of soft growing robots; specifically, the three-dimensional extension of the optimizer designed for planar manipulators. This tool is intended to be used by engineers and robot enthusiasts before manufacturing their robot: it suggests the optimal size of the robot for solving a specific task. The design process models a multi-objective optimization problem to refine a soft manipulator's kinematic chain. Thanks to the novel Rank Partitioning algorithm integrated into Evolutionary Computation (EC) algorithms, this method achieves high precision in reaching targets and is efficient in resource usage. Results show significantly high performance in solving three-dimensional tasks, whereas comparative experiments indicate that the optimizer features robust output when tested with different EC algorithms, particularly genetic algorithms.

Figures (8)

• Figure 1: (a) A soft growing manipulator with continuum links and variable discrete joints do2024stiffness. (b) Sketch of a three-dimensional soft growing robot reaching two targets in an environment with obstacles (side view). (c) A picture of a physical manipulator.
• Figure 2: Degrees of freedom (blue arrows) of the soft growing manipulator addressed in the current work, originally designed by Do et al. do2024stiffness. Each joint can rotate on x and y-axes, growing or retracting on the z-axis. Each joint has its own reference frame, whose z-axis is aligned in the direction of the previous link.
• Figure 3: Example of human-machine interaction with a three-dimensional soft growing robot. A soft manipulator mounted on the ceiling reaches down to help a human load/unload a dishwasher. The human's role is to take dishes from the machine, whereas the robot's role is to safely place them on the shelf.
• Figure 4: (a) Task example: three targets placed on different pillars with a desired orientation. The robot grows from its base $h$ such that its final link is aligned with each target. This task has three targets ($|t|=3$) and three obstacles ($|o|=3$) -- i.e., the pillars where the targets are placed. (b) Optimizer's output example: the solution $\psi$ comprises three configurations $\zeta_{1\rightarrow3}$ that share a unique design $\delta$ (i.e., the three configurations $\zeta$ share the same link lengths). Because the robot may use only a limited amount of links to reach a specific target, depending on its distance to it, each configuration can have a different number of employed links. In this example, the number of links whose endpoint corresponds to the robot's tip are $\bar{n}_1=4$, $\bar{n}_2=5$, $\bar{n}_3=4$.
• Figure 7: The genotype described in Eqn. (\ref{['eq:solution']}) yields the phenotype in the example, with a budget of $n=10$ links. However, only five of them are required to correctly reach the target ($\bar{n} = 5$), an information included in the genotype of Eqn. (\ref{['eq:solution_withExtra']}).