Our locomotion subsystem was fabricated according to our CAD model and it turned out almost exactly the same except for the fact that we had to cut the base plate in half in order to fit it within the machines for fabrication. In addition, we had to raise the base height by 1 inch to make space for a larger and more powerful turntable stepper motor. We are currently utilizing 4 omni wheels that are paired with their individual 45 kg cm DC motors with encoders.
In addition, our electronics fit appropriately within the base according to our design. They include a 12V 15Ah LiFePo4 battery, 3 stepper motor drivers, a 12V to 24V DC to DC converter, a 60A Fuse, 3 12V relays, a 9 DOF IMU, an Intel NUC, a Raspberry Pi 3, a 12V to 5V DC to DC converter, an Arduino Mega, 2 DC motor controllers, and an emergency button.
We had to incorporate ball transfers into our top plate in order to help support our turntable. They were manufactured precisely and implemented according to specifications.
Our Turntable X-Gantry system was also machined and laser cut precisely according to specifications. We are able to translate smoothly along the X direction with a 178.5 oz. in. NEMA 23 stepper motor. Unfortunately, our turntable was initially equipped with the same 178.5 oz. in. NEMA 23 stepper motor; however, it turned out to not be powerful enough to turn the turntable, so we had to switch it out for a 269 oz. in. NEMA 23 stepper motor near the end of our project.
The Z-Gantry was again manufactured according to specifications in our CAD model, and it allows us to translate the base of arm nicely in the vertical direction. However, we did have to switch out the original 68 oz. in. NEMA 17 motor with a 92 oz. in. NEMA 17 motor in order to actuate the Z-Gantry smoothly. In addition, we are using a Microsoft Kinect v2 for the camera.
Our robotic arm is currently built according to our specifications and can properly actuate each of its degree of freedoms. We are using 1 HEBI X5-9 motor and 2 HEBI X5-4 motors for our arm along with a 12V Air pump attached to our granular jammer end-effector.
Our Granular Jammer System is was correctly 3D printed after many iterations. Its flexibility and simplicity has allowed us to easily manipulate the objects at each station. In addition, we are easily able to rotate the end-effector more than 360 degrees because of a revolving tube fitting we bought and integrated into the system.
We successfully utilized our Raspberry Pi Camera Mount with LEDs during the system demos to hold our Raspberry Pi Camera in place and provide it with illumination when looking at horizontally mounted objects.
We have implemented and trained a neural network in Tensorflow that can classify stations and recognize objects, which is shown in the first picture above. The neural network takes in images from our Kinect and outputs class labels, which we then use along with the depth image from the Kinect to determine the locations of the objects. In addition, we are using OpenCV to identify the orientations of objects up close using the Raspberry Pi Camera that is attached near our end-effector.