Design Optimizer for Planar Soft-Growing Robot Manipulators

TL;DR

This work tackles the design optimization of planar soft-growing robots to solve task-specific reaching under obstacles. It introduces rank partitioning to convert a multi-objective design problem into a single objective while integrating obstacle avoidance within the evolutionary operators. The method jointly optimizes reach accuracy, orientation alignment, and manufacturing-effort-related components, yielding more precise, compact, and less undulating designs than prior approaches. Experimental results across diverse tasks show improved inverse kinematics accuracy, reduced resource use, and faster performance, validating the practical value for designing vine-like soft robots. The framework lays groundwork for extensions to 3D environments and alternative optimization strategies, with potential for broader applicability in soft robotics design and manufacturing planning.

Abstract

Soft-growing robots are innovative devices that feature plant-inspired growth to navigate environments. Thanks to their embodied intelligence of adapting to their surroundings and the latest innovation in actuation and manufacturing, it is possible to employ them for specific manipulation tasks. The applications of these devices include exploration of delicate/dangerous environments, manipulation of items, or assistance in domestic environments. This work presents a novel approach for design optimization of soft-growing robots, which will be used prior to manufacturing to suggest engineers -- or robot designer enthusiasts -- the optimal dimension of the robot to be built for solving a specific task. I modeled the design process as a multi-objective optimization problem, in which I optimize the kinematic chain of a soft manipulator to reach targets and avoid unnecessary overuse of material and resources. The method exploits the advantages of population-based optimization algorithms, in particular evolutionary algorithms, to transform the problem from multi-objective into a single-objective thanks to an efficient mathematical formulation, the novel rank-partitioning algorithm, and obstacle avoidance integrated within the optimizer operators. I tested the proposed method on different tasks to access its optimality, which showed significant performance in solving the problem. Finally, comparative experiments showed that the proposed method works better than the one existing in the literature in terms of precision, resource consumption, and run time.

Figures (15)

• Figure 1: Different types of robotic manipulators: (a) a traditional rigid manipulator with localized joints, large body workspace, and significant inertia; (b) a traditional soft robot that lacks dexterity and cannot support significant payloads; and (c) a soft growing manipulator with continuum links, variable stiffness, and variable discrete joints do2020dynamically.
• Figure 2: (a) Sketch of a planar soft-growing robot reaching targets in a maze (top view). The configurations of this solution share the same angle for some of the joints, as the robot progressively grows and retracts to reach the next target. (b) A real prototype of a soft-growing robot.
• Figure 3: (a) Example of a task to be performed by a soft-growing robot: the robot should reach the three targets with a given orientation starting from the home base. The task is defined with three targets ($t=3$) and no obstacles ($o=0$). (b) Example of solution $\psi$ provided by the optimizer. The solution $\psi$ is then composed of three configurations $\zeta_{1-3}$ sharing the same design $\delta$ (the lengths of links are shared among the three configurations). The number of joints can vary within each configuration, as the robot will employ only a sufficient number of links to reach a certain target. In this solution, the joints defining the tip of the robots are $\bar{n}_1=4$, $\bar{n}_2=5$, $\bar{n}_3=5$.
• Figure 4: In (a), the phenotype of a robot generated with the genotype described in (\ref{['eq:solution']}): in this codification, a random robot is generated with $n=20$ links. However, as shown in (b), only seven of them are actually employed to reach the target correctly ($\bar{n} = 7$). Furthermore, the last link is not fully everted (contrary to the internal links composing the rest of the body): this information is codified by extending the genotype as in (\ref{['eq:solution_withExtra1']}). In conclusion, the other thirteen links after the end effector do not need to be included in the design $\delta$.
• Figure 5: Workspace of a robot having $2$ to $5$ links of fixed lengths, where each joint can steer in a range of $\Delta_\theta = [+\pi/6,-\pi/6]$. In the figure, the home base has a fixed orientation, meaning that the first joint cannot steer and the robot everts orthogonally from the base. Any point of the workspace can be reached by the end effector of the robot (with a specific orientation) as the robot can grow and shrink.