-
-
Meet the Team!
-
-
-
-
-
-
-
Abstract
Our product aims to improve existing mechanisms for robot localization. Our product consists of an affordable but accurate system of sensors accompanied with an advanced neural network to perform the robot localization problem. Existing solutions contain 1 of 2 problems: 1. The utilized sensors and technologies fail to output results with sufficient accuracy, or 2. Solutions that predict accurate results are too expensive for commercial use. With a combination of neural network computing and optimized sensor technology and placement, we aim to innovate upon existing solutions to create a more seamless and accurate system. We believe this project can have a wide impact, from large corporations and warehouses to universities to individual students, and advance the current field of robotics.
What it does
We use sophisticated Ultra Wideband sensors to design and implement a detection system for robots throughout a designated testing area. Accurate data will be collected in real-time and sent wirelessly to a software API, where we run a modular multiple perceptron neural network and output a predicted location. The overall goal is to use different technology combinations and improve upon accuracy results within existing solutions.
How we built it
Our project consists of four UWB modules (anchors) that send a signal to additional UWB modules connected to the robots (tags). The tags report the signals they receive from the anchors to our server via MQTT protocol, and a host computer listens to the server and receives the data remotely. The information is processed through a neural network, whose results are outputted in real-time to our website through an HTTP post, where students can track their robots in a 2D map of the testing site.
Challenges we ran into
Our initial project idea contained a type of smart outlet system, but we quickly realized it would not be sufficiently innovative upon existing solutions. Consequently, we managed a hard pivot about halfway through the semester to tackle a larger problem in robot localization. Our biggest challenge thus far - aside from racing against time - has been setting up our hardware, since our project was initially Bluetooth Low Energy-based, but we had to switch to UWB. Due to the current worldwide chip shortage, our progress was further delayed. Additionally, we initially transferred our information from the tags to host comptuer via UDP protocol, but we found the protocol to be extremelly buggy, so we changed our code to transfer information to a remote server via MQTT instead.
What's next for Team 22- RoboLoc
The next steps for our team are to further train our Neural Network with more randomly collected data points, as well as to experiment with different normalization features and learning functions in order to optimize the accuracy of our results. Currently, we are using a public server, ideally we would have a private server so that the data cannot be accessed by anyone. Lastly, we would like to tweak the transmission frequency of our UWB modules in order to avoid interference within the system, since our results are not accurate when there are more than one active tag.
Log in or sign up for Devpost to join the conversation.