**6. Conclusions**

This work presents a novel approach to soft robotics with the design of a flexible and compact soft joint. It is not only a low-cost prototype, assembled by 3D printing. It also has a morphology that allows better handling of external loads and gravity thanks to its blocking configuration. Actuated by tendons, the proposed design has a morphology with two main configurations of flexion, which provides more versatility and a flexion limit, unlike previous designs. These characteristics and configurations can be modified through the parameters of the joint morphology, to achieve different fields of work and functionality.

A mathematical model of the inverse kinematics of the soft joint is also presented to obtain the length of the tendons as a function of the morphology and the position (orientation and inclination) of the end of the joint. The modeling of the soft morphology is a complex task, but a simplified and sufficiently accurate kinematic model has been shown. For its validation, the soft link prototype has been built and simulation and experimental studies have been carried out.

According to the capabilities of the solution described and demonstrated throughout the paper, the soft joint proposed in this work shows an improvement over other designs and it could be used for many different applications requiring manipulation of loads. Our main application will be the use of this joint as an arm for the humanoid robot TEO so that the robot can perform manipulation tasks with the use of a gripper connected to the arm tip. There are several uncertainties and mismatches that affect the model of the prototype, especially when this is a low-cost 3D printed solution. For instance, the curvature of the real model is not constant, the tension and length of tendons are not exact, and small variations in the radius of the winches happen after several turns. Despite these facts, the proposed model is accurate enough to represent the kinematics of the system and will allow a later control of the soft joint in closed loop. Further research will lead to reducing these inaccuracies and prototyping effects and to closing the control loop and testing the platform with different loads during manipulation interactions.
