Page 71 - IJAMD-1-1

International Journal of AI
for Material and Design Machine learning for gripper state prediction

characteristics of thermoplastic polymers, particularly their actuation. The relationship between these factors is pivotal;
capacity to soften under heat without breaking and then a joint with high stiffness counteracts bending with a
harden as they cool. For the backbone and joints, c-PLA is greater moment, while a joint with lower stiffness offers
selected for its electrically conductive properties. Stiffness minimal resistance, allowing for a wider range of motion.
modulation in the joints is accomplished through joule Crucially, the stiffness of each joint is temperature
heating, a technique where passing an electrical current dependent, and the capability of adjusting this
through a resistive material generates heat. In this design, variable enables a virtually unlimited array of stiffness
the carbon-infused c-PLA joints are heated by applying configurations. This, in turn, translates into a myriad of
electricity to each joint’s terminals. This heating reduces the potential shapes and positions for the gripper, as illustrated
joint’s stiffness, making it more pliable, the process reversed in Figure 1E, each suited to different tasks and objects. The
once the temperature falls back below the polymer’s glass precise control of this variable is what imparts versatility
transition point, as demonstrated in Figure 1C. The multi- to the gripper. However, the sheer number of potential
nozzle 3D printing process of the multimaterial gripper is stiffness configurations presents a computational challenge.
depicted in Figure 1D. By selectively heating various joints,
different gripping configurations can be attained, enabling Simulating and analyzing every possible combination of
the gripper to conform to a range of geometrical shapes joint stiffnesses and the resulting shapes of the gripper
(Figure 1E). would be infeasibly resource intensive. To address this,
we integrated machine learning algorithms as a means to
2.2. Development of a predictive model for the circumvent the need for exhaustive numerical simulations,
gripper with variable joint stiffness providing a more resource-efficient approach to model the
gripper’s behavior.
In this work, we investigated the dynamic behavior and
kinematic properties of a robotic gripper finger articulated Figure 2 outlines the systematic approach employed
at two joints. The primary focus was to understand how in this study. It commenced with the initial step of
the bending angles at these joints are influenced by two numerical modeling and simulation to identify critical
main factors: the distance at which the actuation cable is input parameters, such as the pull distance and the ratio
pulled and the instantaneous stiffness of the joints during of stiffness between the joints. In addition, this phase

Figure 2. Flowchart illustrating the methodology of machine learning application in the actuation control of a soft robotic gripper.
Abbreviation: KNN: K-nearest neighbor regression.

Volume 1 Issue 1 (2024) 65 https://doi.org/10.36922/ijamd.2328