The McKibben muscle belongs to the type of muscles known as braided muscles. It is made of an inner hyper-elastic tube, surrounded by a braided shell made of inextensible threads; both ends provide mechanical and pneumatic seal. A finite element model of a McKibben pneumatic muscle was built and experimentally validated. The model is based on characteristic parameters of McKibben muscles. It takes into account the non-linearity of the constitutive material of the inner tube. It does not simulate backslashes between the tube and the shell at rest condition, but it models threads and rubber that are always connected. However, it does not consider friction among threads. In order to build and to validate the proposed numerical model, an experimental prototype of the muscle was designed and built. Both isotonic and isometric tests were carried out. Same tests were simulated in the finite element environment. The model validation was performed by comparison between experimental and numerical results.

In collaborative robotics, to improve human–robot interaction (HRI), it is necessary to avoid accidental impacts. In this direction, several works reported how to modify the trajectories of collaborative robots (cobots), monitoring the operator’s position in the cobot workspace by industrial safety devices, cameras, or wearable tracking devices. The detection of the emotional state of the operator could further prevent possible dangerous situations. This work aimed to increase the predictability of anomalous behavior on the part of human operators by the implementation of emotional intelligence (EI) that allows a cobot to detect the operator’s Level of Attention (LoA), implicitly associated with the emotional state, and to decide the safest trajectory to complete a task. Consequently, the operator is induced to pay due attention, the safety rate of the HRI is improved, and the cobot downtime is reduced. The approach was based on a vision transformer (ViT) architecture trained and validated by the Level of Attention Dataset (LoAD), the ad hoc dataset created and developed on facial expressions and hand gestures. ViT was integrated into a digital twin of the Omron TM5-700 cobot, suitably developed within this project, and the effectiveness of the EI was tested on a pick-and-place task. Then, the proposed approach was experimentally validated with the physical cobot. The results of the simulation and experimentation showed that the goal of the work was achieved and the decision-making process can be successfully integrated into existing robot control strategies.

In this article, we present, for the first time, a soft robot control system (SofToss) capable of throwing life-size objects toward target positions. SofToss is an open-loop controller based on deep reinforcement learning (RL) that generates, given the target position, an actuation pattern for the tossing task. To deal with the high nonlinearity of the dynamics of soft robots, we deploy a neural network to learn the relationship between the actuation pattern and the target landing position, i.e., the direct model (DM) of the task. Then, an RL method is used to predict the actuation pattern given the goal position. The proposed controller was tested on a modular soft robotic arm, I-Support, by tossing four objects of different shapes and weights in 140-mm squared target boxes. We registered a success rate of almost 65% of the throws in two actuation modalities (i.e., partial, keeping one module of the soft arm passive, and complete, with both modules active). This performance raises to 85% if one can choose the number of modules to actuate for each throwing direction. Furthermore, the results show that the proposed learning-based, real-time controller achieves a performance comparable to that of an optimization-based, nonreal-time controller. Our study contributes to the foundations for bringing soft robots into everyday life and industry by performing more complex, dynamic tasks.

The study of bio-inspired structures and their reproduction has always fascinated humans. The advent of soft robotics, thanks to soft materials, has enabled considerable progress in this field. Over the years, polyps, worms, cockroaches, jellyfish, and multiple anthropomorphic structures such as hands or limbs have been reproduced. These structures have often been used for gripping and handling delicate objects or those with complex unknown a priori shapes. Several studies have also been conducted on grippers inspired by the seahorse tail. In this paper, a novel biomimetic soft pneumatic actuator inspired by the tail of the seahorse Hippocampus reidi is presented. The actuator has been developed to make a leg to sustain a multi-legged robot. The prototyping of the actuator was possible by combining a 3D-printed reinforcement in thermoplastic polyurethane, mimicking the skeletal apparatus, within a silicone rubber structure, replicating the functions of the external epithelial tissue. The latter has an internal channel for pneumatic actuation that acts as the inner muscle. The study on the anatomy and kinematic behaviour of the seahorse tail suggested the mechanical design of the actuator. Through a test campaign, the actuator prototype was characterized by isotonic tests with an external null load, isometric tests, and activation/deactivation times. Specifically, the full actuator distension of 154.5 mm occurs at 1.8 bar, exerting a maximum force of 11.9 N, with an activation and deactivation time of 74.9 and 94.5 ms, respectively.