This is a guest contribution by Egor Driagin, CMO at Top3DGroup
A team of Austrian and German scientists decided to reproduce an elephant trunk. They made a conceptual demonstration model by 3D printing the majority of parts and connecting a wrestling robotic manipulator to a spiking neural network (SNN). This article will cover how it turned out to be.
This project belongs to a team of scientists from two universities: the University of Tübingen and Graz University of Technology. Their solution can be described with the word ‘over-engineering’, since usually producing flexible robotic arms means using elastic components with pneumatic or hydraulic drives but in this case the construction is rigid.
The overall concept of the mechanical part is pretty simple: the components of the trunk are connected with the ball joints and their height and angle can be regulated using three or four rack and pinion actuators. The latter feature individual servomotors. In theory, it’s possible to increase the length and degrees of freedom for as long as one wants, but in practice, the size of the construction is limited by the power of servomotors, the strength of supporting parts, and computing power. The majority of parts of the model are made of PLA using a regular desktop FDM 3D printer with the Bowden-style system. An elastic polyurethane grip was produced on a separate 3D printer with a direct style extruder. The algorithms are the most complex things about this project.
The system is controlled by a self-learning AI based on the spiking neural network. It tracks the way the motors work, compares the motions to the set positions, and corrects the servomotors, which makes the near-millimeter level of accuracy possible. Data is processed in different ways: in the version with four-actuator segments, the network receives normalized values of the angles by the X and Y axes and elongation by the Z-axis, while the three-actuator version operates with the data that is received directly from the rack and pinion actuators.
Not only the developers showed how to produce inexpensive trunk-like robotic systems using basic 3D printing equipment, but also demonstrated how such robots can work with impulsive neural networks based on modern architectures. According to them, the plan is to ideally make a system that is constantly learning, where a robot would be launched without any data and will try to complete the tasks, generating exercises during the self-learning process.
The intention of the developers is that such trunk-like robots will become an alternative to common industrial multiaxial robots and will be used for the same purposes: for example, automotive production lines or to assemble electronic devices. The possibility to integrate sensors and algorithms to avoid obstacles is considered. Such a thing would help to improve safety while working with people. Another possible solution: using snake-like robots in search and rescue operations.