Festo's BionicSoftHand is the latest robotic hand attempting to immitate the way a human hand works. Using artificial intelligence, the hand is able to use its fingers to work out how to manipulate an object by gripping and moving it around.
The BionicSoftHand relaxing an object on its hand before manipulating it through movements. (Image Credit: Festo)
The robotic hand learns through positive reinforcement and instead of imitating a specified action, the hand is given a goal. All of this is done through trial and error and relies on positive and negative feedback it receives to improve its actions until the task is completely solved. It also can rotate certain objects, more specifically, a 12-sided cube, with the opposing points on the cube being turned over on the opposite side. This is taught to the robotic hand in a virtual environment with a digital version of the hand, created from data of a depth-sensing camera and artificial intelligence algorithms.
The digital model of the hand allows it to be trained much faster, especially if there is more than one hand that needs to be trained. In massive learning systems, the knowledge is shared with all virtual hands, and they work with the new knowledge, only making one mistake. After the tasks have been successfully carried out, the solution becomes available to all models. Once all virtual hands have been trained with no mistakes, the data is transferred to the BionicSoftHand. The newly learned virtual concept allows the hand to turn any object over to any side and move them around.
The robotic hand, both the real model and digital model go through tests until a solution can be reached. (Image Credit: Festo)
There is no skeletal structure in the BionicSoftHand, which means it doesn't have any bones like the human hand does. Movements are controlled through the pneumatic bellows structures in its gripper fingers. It works in two ways, both of which control the flexibility and stillness of the fingers. The first way is when the chambers are filled with air, causing the fingers to bend and when the chambers are empty, the fingers are set in a relaxed position. Both thumb and index finger have a swivel module, allowing the two gripper fingers to be moved around, giving the hand twelve degrees of movement. All the bellows in the gripper fingers are enveloped in a 3D textile cover made of elastic and high strength fibers. What this means is the textile can be used to determine where the structure expands, where it's creating force and where it's no longer expanding
A small, digitally controlled valve terminal designed by developers was mounted above the hand to minimize tubing. This makes it so that the tubes that control the fingers do not have to be pulled through the arm. This allows the hand to be connected and operated with one tube for supply and exhaust air. Precision with the gripper fingers can be controlled with the proportional piezo valves found in the wrist.
The elastic materials, pneumatic kinematics and lightweight components makes the hand unique from ordinary electric or cable-operated robot hands, making it a lot cheaper to create. Gripper variants with three or four gripper fingers are also a possibility due to their design.
Coupled with lightweight pneumatic robots like the BionicCobot or the BionicSoftArm, collaborations with humans is possible in a safe workplace environment. They do not require any safety measures to take place like factory robots do. This enables the BionicSoftHand to be used in the workplace in the future - especially in factories. Due to its strength and high sensitivity, the hand can be used in assembly line workspaces and can be used in service robotics.
Have a story tip? Message me at: cabe(at)element14(dot)com
http://twitter.com/Cabe_Atwell
