"Be so busy chasing your goals that you have no time for criticism."
An adaptive admittance controller for collaborative drilling with a robot based on sub-task classification via deep learning
B. Guler, P. Niaz, A. Madani, Y. Aydin, C. Basdogan
This paper is the result of my M.Sc. work. An adaptive admittance controller was designed such that a user interacting with a collaborative robot would be able to smoothly move the robot towards a target, and then interact with the target stably and safely without compromising stability. The adaptive controller chose a low damping value for the so-called Driving phase so that human effort would be low during that time, and a high damping value for the Contract phase, so that stability would be guaranteed as long as the robot is in contact with a stiff environment. The system would use a time-series classification deep learning model in order to determine which subtask (phase) of the task the user is in at the moment (this is how human intention is modeled and detected) and adapts the admittance controller's damping value accordingly. I was the first author of the paper (see this revision) until the final revision, where extensive work was done for the stability analysis due to the request of the reviewers, which was outside the scope of my thesis.
Robot-Assisted Drilling on Curved Surfaces with Haptic Guidance under Adaptive Admittance Control
A. Madani, P. Niaz, B. Guler, Y. Aydin, C. Basdogan
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022, accepted)
In this paper, we used a 6-DOF adaptive admittance controller to perform drilling on a curved surface with an unknown surface geometry via a collaborative robot, at custom non-perpendicular drilling angles, with the help of 3D scanning done with a Microsoft Kinect camera. A Microsoft Hololens AR Goggle guided the user throughout the process, increasing task efficiency further. Once target position and surface normal information were extracted via 3D scanning (using the Kinect camera) followed by geometric referencing, they were sent to the robot. The human then guided the robot through the obstacles and convex/concave areas of the curved workpiece towards the target. Once the robot was close enough to the target, a so-called "haptic guidance module" took over, aligning the drill bit tip exactly with the target, at the chosen drilling angle (specified by azimuth and polar angles relative to the workpiece). The admittance controller was locked on the chosen drilling vector, effectively becoming a 1D admittance controller along the drilling angle. The operator then took back control of the robot and drilled through the workpiece by simply pushing the robot forwards. The adaptive controller selected low damping during driving the robot and increased the damping to a safer value when the robot was close enough to the target. Much higher damping was chosen for the final drilling phase to maximize stability and accuracy during the drilling.
Developing an adaptable pipe inspection robot using shape memory alloy actuators
A. Hadi, A. Hassani, K. Alipour, R.A. Moghaddam, P. P. Niaz
Journal of Intelligent Material Systems and Structures
In this paper, we developed a pipe inspection robot (a.k.a. crawler) actuated partially by shape memory alloys replacing the bulky hydraulic actuation mechanism of typical pipe inspection robots. Using this actuation mechanism, the robot can climb through vertical, bent, and slippery pipes without having to rely on bulky and expensive hydraulically actuated systems to stay stable in vertical or slippery pipes. The SMA actuators were much smaller than their hydraulic counterparts in the industry, and required much less electrical energy to be actuated, increasing efficiency dramatically.