Domain Randomization and Model Order Reduction of Soft Robotic Crawler (Ongoing)
“All models are wrong, but some are useful.” — George Box
Domain randomization is a popular technique for computer scientists to make models robust to environmental change. From a mechanical engineering perspective, domain randomization could possibly make a less accurate model achieve similar level of accuracy with a more accurate model by exploring more environments.
Thesis research question: How to do maintain peak performance, while ensuring the robot is more robust to environmental change.
McGill Summer Undergraduate Research (SURE) 2022
Robots made up of soft and compliant actuators and materials are naturally easier to accomplish tasks with simpler designs and more cost effective compared to rigid robots. They are also safer for applications which requires human interactions. All of those properties made soft actuators a great application for grasping tasks. They are natrually compliant and will fit to the shape of object making the grasping tasks easy to be accomplished. Fragile objects can be grapsed with ease using soft actuators as well.
However, soft actuators are very hard to have accurate enough analyitcal models established, design of soft-robots were highly dependent on trial-and-error processes which is time consuming.
Therefore I have explored and created a co-optimization design space between morphology and control for soft-robot grasper and created simulation scenes in SOFA framework with a simple grasper and controller for model order reduction.
Honours Design (Mech 494)
In this course, I designed the baseline grasper introduced before and caded out all the components ensuring the proper operation of the grasper.
