Project
# | Title | Team Members | TA | Documents | Sponsor |
---|---|---|---|---|---|
71 | Extend IMU Degrees of Freedom for Pose Estimation Using AI on Chip |
Chirag Rastogi Lukas Zscherpel |
Yixuan Wang | design_document1.pdf design_document2.pdf final_paper1.pdf presentation1.pdf proposal1.pdf |
|
EXTEND IMU DEGREES OF FREEDOM FOR POSE ESTIMATION USING AI ON CHIP Team Members: - Chirag Rastogi (chiragr2) - Lukas Zscherpel (lukasez2) # Problem An Inertial measurement unit (IMU) is a combination of sensors that collects data based on movement. IMU’s normally include an accelerometer and a gyroscope which track the specific acceleration and the angular acceleration of the object. The sensors are: Accelerometers: Used to measure linear acceleration in three dimensions. This information can be used to estimate the velocity and position of the object over time. Gyroscopes: Used to measure angular velocity in three dimensions. This information can be used to estimate the orientation of the object over time. Magnetometers: Used to measure the direction of the Earth's magnetic field. This information can be used to determine the orientation of the object with respect to the Earth's magnetic field, which can be used to correct errors in the orientation estimate obtained from the gyroscopes. IMU’s are used in a wide range of applications but they are really important in the medical field and in consumer electronics. Some example applications include movement tracking on patients to detect disorders or even tracking movement in your cell phone to get its orientation. 9DOF IMU sensors can be found for as low as $10-$20 for basic models, but these sensors have lower accuracy. For projects that require greater accuracy, the cost can go upto 300$ (https://x-io.co.uk/ngimu/) and this limits projects that require multiple such devices. # Solution An AI on chip solution may have the potential to reduce the cost of 9DOF IMU sensors by enabling the integration of multiple sensors and processing functions onto a single chip, which can simplify the design, reduce the bill of materials, and lower the manufacturing costs. By leveraging AI algorithms among others, an AI on chip can enable 9DOF IMU sensors to perform advanced sensing and processing tasks on-device, reducing the data transmission requirements and minimizing the need for external computing resources. Our solution is to take a cheap 6 DOF IMU and combine it with a RNN that we train to calculate the other 3 DOF that a magnetometer normally provides. We will then take this AI model and put it onto a chip. The AI on chip will work together with the 6DOF IMU to emulate a 9 DOF IMU in a handheld format. # Solution Components ## Subsystem 1: Inertial Measurement Unit This subsystem will be an 6 DOF IMU that we acquire from a third party distributor. We will have to research what the IMU will output and how to connect to it as well as how to calibrate the IMU. We are considering using an Adafruit ISM330DHCX as the IMU ($20) and the MPU-6050 (3$). https://www.adafruit.com/product/4502 https://www.amazon.com/HiLetgo-MPU-6050-Accelerometer-Gyroscope-Converter/dp/B01DK83ZYQ?th=1 ## Subsystem 2: Control System We will have a control system (microcontroller) that is designed by a student that will process the data outputted by the IMU and provide it to the AI on chip. It will then take the output of the AI model along with the other data and output it to the usb port. We are considering using an ESP32 microcontroller for this subsystem. ## Subsystem 3: AI on Chip AI on chip either through Nvidia Jetson or fpga that will take the output of the IMU and predict what the orientation of the device will be. The model will be created and trained on a students laptop on data acquired. The model will then be fitted and tuned to fit onto the processor that we choose https://ieee-dataport.org/open-access/estimating-relative-angle-between-two-6-axis-inertial-measurement-units-imus. ## Subsystem 4: PCB and Power Supply For our project we will mount everything to a PCB that we design. The pcb will host all of the other subsystems as well as a USB interface that will provide power as well as output the data to an external source such as a laptop to be recorded. # Criterion For Success The output of the 6 DOF imu is displayed and recorded on a separate computer. The calculated 3 DOF are displayed and recorded on a separate computer. The PCB including the IMU is able to be turned off and disconnected from a computer. |