Viscous, Soft Robotic Actuators
Skills:
-
CAD (Fusion 360)
-
3D Printing/Rapid Prototyping
-
Dynamixel Servos
-
FEA
-
Beam Theory
-
Fluid Mechanics
Overview
Soft robotics have been an rapidly-growing area in the field of robotics, with many applications in medical devices, wearables, and prosthetics. Fluidic elastomeric actuators (FEAs) are a special soft robotic actuator which are generally driven through a changing pressure distribution within the body of the actuators. They are prized for their simplicity, intuitive driving mechanism, and infinite passive degrees of freedom. However, one limitation of these systems is an exponential scaling in complexity with an increasing number of motions. Thus, the lab developed these novel FEAs which can be driven by a single pressure inlet and can achieve a complex range of motions. The novelty behind these actuators are the bellows combined by thin tubes, which induce viscosity into the system. This generates a temporary non-uniform pressure distribution within the actuators, allowing for more complex motions while only requiring one inlet/outlet.
A video demonstrating a previous paper where they created a walking robot using such actuators
Background
The lab had already performed previous work with these FEAs. One huge benefit of their previous work was they had made a COMSOL model which could predict the transient motions. However, there was still improvements to be made. First, non-uniform pressure distributions within the actuator were only transient still and the actuator could not hold one of these poses. Secondly, the ability for the FEAs to only linearly extend had not been explored yet. Thus, our job was to implement these improvements and find another application for the actuators. We decided, for our project, to create a robotic gripper which could grip, scroll, and click a mouse wheel. We also improved the COMSOL model to better predict motions based on pressure inputs.
A CAD rendering of the actuators. By inflating/deflating the actuator at the inlet, one side changes faster and causes a bending motion
Project
I worked with two other MEng students on developing these actuators. After we determined what we wanted to do with the actuators, create the scrolling robot, we then needed to figure out how to maintain a non-uniform pressure distribution within the actuator. We came up with the idea of incorporating a valve in-between the two columns of bellows so the two columns could maintain their own pressure separately. The valve would also allow us to easily control flow between the two columns.
After we built these new FEAs, we had to design a new mount for the FEAs. One of the other students created the initial design, while I created the final design in Autodesk Fusion 360. We then 3D printed them on a Prusa i3. Unlike the old walking robot, we moved the syringe pump which drove the actuators to be further away from the actuators as we did not need the entire setup to move. This also proves these actuators can easily be driven from a central platform, proving their versatility and simplicity of the driving mechanism.
After everything was physically built, it was time to write a new control algorithm for our new robotic gripper. I was the person with the most robotics/electronics experience so this was my main job. The previous people had used an older version of the servos and we had ordered new servo motors which could move faster and had more torque. However, this meant writing up an entirely new algorithm to control the motion of the FEAs.
Because we also needed precise speed, position feedback, and extremely high torque, we were using Dynamixel servos, which unfortunately were more complicated to wire and control than a typical servo motor. The previous microcontrollers were also lost so the previous code did not work; meaning I had to figure out how to actually communicate with the Dynamixel servos from scratch.
While writing the control algorithm, I also had to figure out how the speeds at which to inflate the arms to achieve a linear motion with a constant pressure profile inside vs. a transient, non-reciprocal motion with a temporal non-uniform pressure profile. After I had written the control algorithm and figured out the required motions (click, scroll, gripping with the valve, etc.), we could finally record the video demo, perform testing, gather data for the predictive model, and refine the control algorithm.
I performed all of the data collection as well, which included wiring up different sensors such as an absolute pressure sensor to a separate microcontroller and collecting data asynchronously. Our work over the two semesters culminated in a paper submission to the RoboSoft 2024 Conference and was just recently accepter for publication!
