Publication Date

Fall 2023

Document Type

Capstone Project

Degree Name

Master of Science


Computer Science

First Advisor

Xueqing Tang

Second Advisor

Yunchuan Liu

Third Advisor

Steve Shih


Robotics is a multidisciplinary field that involves the design, construction, programming, and operation of robots. Robots are mechanical or electromechanical devices that are outfitted with sensors, processors, and actuators to carry out tasks autonomously or under the direction of a human. Robotics offers several benefits and uses in a variety of fields and industries. Robots can carry out repetitive activities with great accuracy and reliability. They can produce products with consistency in quality by performing operations with extreme accuracy.

Our anticipated approach merges traditional methodologies with forward-thinking innovations. Central to our design philosophy is the integration of tools such as Android Studio, GitHub, Git, and the Java programming language. We propose a robot with a constrained starting volume of 18x18x18 inches and a material flexibility of up to 0.25 inches. Integral to its design will be a web camera to enhance navigation capabilities, with the inclusion of April Tags for a comprehensive understanding of the game field's layout. The control system, a cornerstone of our design, is expected to feature an Android device interlinked with two team controllers, a driver station, and a Wi-Fi-enabled robot controller, with potential interfacing through the REV Robotics Expansion Hub or the REV Robotics Control Hub.

Our strategy envisions a robot capable of navigating the competition's multifaceted challenges, including a 30-second autonomous period, a two-minute driver-operated phase, and a climactic 30-second endgame. For robot manipulation learning, we provide a vision-based architectural search method that identifies relationships between high-dimensional visual inputs and low-dimensional action inputs. Our method automatically creates structures as it learns the task, coming up with fresh ways to associate and attend picture feature representations with actions and features from earlier levels. Comparing the obtained new architectures to a current, high performing baseline, they show superior task success rates, sometimes by a significant margin.