Generative Robot Design: Morphological Computation & Underactuated Grippers

Table of Links

Abstract and I. Introduction

II. Methods

III. Generation of Tendon Driven Grippers

IV. Results and Discussion

V. Conclusion and References

Abstract— Robots’ behavior and performance are determined both by hardware and software. The design process of robotic systems is a complex journey that involves multiple phases. Throughout this process, the aim is to tackle various criteria simultaneously, even though they often contradict each other. The ultimate goal is to uncover the optimal solution that resolves these conflicting factors. Generative, computation or automatic designs are the paradigms aimed at accelerating the whole design process. Within this paper we propose a design methodology to generate linkage mechanisms for robots with morphological computation. We use a graph grammar and a heuristic search algorithm to create robot mechanism graphs that are converted into simulation models for testing the design output. To verify the design methodology we have applied it to a relatively simple quasi-static problem of object grasping. Designing a fully actuated gripper may seem simple, but we found a way to automatically design an underactuated tendondriven gripper that can grasp a wide range of objects. This is possible because of its structure, not because of sophisticated planning or learning. To test the applicability of the proposed method in real engineering practice, we used it to create physical prototypes. Simulation results together with results of testing of physical prototypes are given at the end of the paper. The framework is open source and the link to GitHub is given in the paper.

I. INTRODUCTION

Designing robots is a multiphase process aimed at solving a multi-criteria optimization problem to find the best possible detailed design. When it comes to creating something new, the possibilities are endless. Creator must consider factors like shape, mechanics, materials, sensors, controllers, and etc. to bring design to life. The solution space knows no bounds, making each design choice a crucial step towards bringing the vision to reality. When the designing process is manual, it is difficult to prove the optimal solution numerically because it relies on the designer’s experience and engineering intuition [1].

A. Contribution

This research output is based on our previous studies on morphological computation as [2]–[4]. As a step towards, we developed the “rostok”[1] framework for generative and interactive design of linkage mechanisms. The proposed framework is general enough and an advanced user can use it for a linkage synthesis for any robotic purposes: robot arms, legs, fingers etc.

To verify applicability of the methodology for real engineering tasks, we have applied it to solve the task of computational design of underactuated tendon driven grippers, as a relatively simple quasi-static task. Fig. 1 shows steps of the design process: (1) a heuristic algorithm generates a graph; (2) the graph is converted into a simulation model; (3) virtual experiments are conducted using the simulation model, and the results are evaluated using a reward function; (4) the reward value obtained is used to guide the exploration of the design space; (5) the final design options are verified by means of physical prototypes.

For the sake of clarity, the contribution of the paper consists of several items:

• Combining morphological computation paradigm with generative design, the paper proposes an approach to the automatic generation of linkage mechanisms with “mechanical intelligence”. The performance of these mechanisms is mainly determined by their mechanics, or morphology, rather than sophisticated trajectory planning and control.

• We successfully implemented the graph grammar approach to explore the design space of anthropomorphic underactuated tendon-driven grippers. Grasping is done by means of gripper morphology, rather than sophisticated control strategies.

• To ensure the highest performance of generated grippers, physical prototypes were constructed to validate both the solution and the overall methodology. This step aimed to assess the practicality and effectiveness of the methodology in real-world engineering applications.

B. Related Work

Model-based design optimization approaches can be used to efficiently synthesize linkage mechanisms together with needed control. Gradient-based together with global optimization algorithms are used to optimize geometry, mass distribution, or actuation within specific boundaries [5]–[7]. Properly formalized optimization tasks allow to efficiently find suitable solutions, however model-based design optimization approaches strongly depend on user inputs.

As an alternative approach, the generative design needs minimum input user to create unique and unexpected solutions. However, the lack of the current simulation software the evolutionary design creates highly abstract structures that hardly can be physically reproduced [8]–[10].

GD is intended to automatize co-optimization of mechanical structure and joints trajectories to satisfy poorly formalized tasks from a user. Generative adversarial networks (GANs) [11], graph neural networks (GNNs) [12], [13], and deep neural networks (DNNs) [14] are used to generate robots for a wide range of scenarios. The design of the robot is defined by a graph. Recursive grammar rules for changes are applied over the graph. The approach allows the search space to be modified by changing the set of rules.

A modular co-evolution strategy been implemented in [15] where primitive agents dynamically self-assemble into complex bodies and learn to coordinate their actions. The approach outperformed static methods in simulated environments, showcasing improved adaptability to environmental changes and agent structure variations.

An innovative approach to robotic manipulator design, offering a streamlined pipeline that enables rapid creation and customization of manipulators with knitted, glove-like tactile sensors is given in [16]. By applying modular components and predefined rules, engineers can quickly prototype manipulator designs, fine-tune their shape, and seamlessly integrate tactile sensing, with successful real-world applications demonstrated.

A novel underactuated compliant hand emulator, leveraging advanced simulation software and the adaptive synergy concept to simplify mechanical and control complexity, is introduced in [17].

[1] https://github.com/aimclub/rostok