Design

DESIGN CRITERIA & Desired Functionality

In design, we look for how original or ingenious our project is as our main criteria.

Uniquely to us, we hope to utilize one arm to do the necessary folding actions. Other towel folding robots tend to use both hands to grab and fold, but we hope to use robotic arms more efficiently by using only one arm. We also aim to use a camera that is inherent to the robot, to streamline efficiency in order to avoid setup time and costs for external cameras.

This project aims to accomplish a folding task, which is previously done on much more precise and advanced robots, on a robot that is much simpler. We hope to achieve results without having precision grippers and thus allowing it to be a more widespread and reusable technique.

Basic DESIGN overview

Left hand camera for image taking
Baxter left arm and gripper to maneuver the towel
Using CNN and homography to detect corners of towels

Design Choices

Machine Learning Algorithm

For more information on the sensing algorithms used in this project, visit the Implementation page.

Camera

During our design process, we considered several cameras before settling on our final implementation. Ultimately, we went with the Left Camera due to a number of technical limitations with the Head Camera and Right Camera. By using the Left Camera, we were able to create an all-inclusive system that does not require additional setup with an external camera, streamlining the efficiency of the entire process. A summary of our reasoning can be found below.

Head Camera

Right Camera

Left Camera

USB Camera

Gripper

We experimented with a variety of grippers throughout the duration of our project. Initially, we wanted to use a gripper with longer pincers in order to grab more of the towel fabric. Then, we transitioned to a shorter pincer in order to retain greater control and precision of the end-effector. We even considered 3D-printing our own gripper because there are several members of our team with CAD experience. Finally, specifically for our earlier trials and test-runs, we used a modified gripper with foam or Silicone tips so that the gripper could better hold onto the towel fabric.

Ultimately, we found that the only requirement necessary was that our gripper was able to close all the way. This required a gripper with a shorter width piece such that the Baxter arm could fully close the pincers. This is the only requirement for our project works, meaning that our project is robust enough to support almost every type of gripper available to Baxter.

Below are two photos of grippers we used in our project. Note that both have the small width portion necessary for the gripper to close fully.

Short gripper

with Silicone Padding

long gripper

Angle of Attack

Furthermore, another design decision we needed to make was with regards to the angle of attack of the gripper. Initially, we wanted to have the gripper have a more horizontal angle of attack approaching the towel from the side. This way, we would be able to grab more of the towel with length of the gripper providing a more secure grip. However, as we were experimenting with the gripper, we found that using such an angle of attack would be infeasible as the open position of the gripper at a tilted or non-vertical was too wide such that one pincer of the gripper would collide with the table; therefore, when we closed the gripper it would not catch the towel at all. To mitigate this, we decided to go with a vertical angle of attack so that the gripper would grab the table it head-on. This would also give us flexibility in allowing us to grab any point on the table in order to execute the fold, as opposed to a tilted or horizontal angle of attack which would have to be executed at the edge of the towel.

Because a vertical angle of attack would require our gripper to just use the tip in order to grab the towel, this meant that the sensing, planning, and actuation components associated with recognizing the towel must be very precise. This is because it is in general very difficult to fold towels and other articles of clothing because the fabric is only a few millimeters thick. The fact that our implementation works supports the overall robustness of the rest of our system, from the sensing of the corner of the towels to the planning of the movements to the execution of the fold, in order for it to successfully grab onto such a small piece of fabric with little margin for error.

Tilted Angle of Attack

Vertical angle of attack

Folding Sequence

For our project, we wanted to do something different from research projects from which we drew inspiration. Instead of the two-arm approach that was used in some other popular clothes-folding videos, we ended up using a one-arm approach. This posed a significant challenge because we would not have another arm to stabilize the towel; instead, we would need to rely on the precision and accuracy of our single gripper in order to properly execute the fold. In order to use one arm, we decided to use a unique fold for our demonstration that could successfully be executed with one gripper alone. Much like folding a napkin, hand towel, or American flag, Baxter folded our towel using a triangular fold, sweeping from one corner to the opposite.

HoW DO These Design Choices Impact How Well The Project Meets Design CriteriA That would be encountered in real engineering application, such as robustness, durability, and efficiency?

As mentioned above, our decisions in the camera, gripper and and angle of attack in approaching the clothing item allows us to accomplish our goals while considering robustness, durabiltiy, and efficiency in real life scenarios. For the camera, we have been able to streamline the efficiency by using a camera of the same frame as the gripper, to minimize calculations. Since the only requirement necessary is for our grippers to close all the way, this meant that our project is robust and can support almost every type of gripper available to Baxter. Our angle of attack allows the robot to grab the corners in a variety of positions, and still success in the movement sequence, thus the lack of need of perfection leads to increased robustness and efficiency.

Page updated

Report abuse