Abstract
dc:description.abstractThis paper presents a solution for determining if a wheeled robot is stuck that is operating without high fidelity localization sensors or operating in a sensor denied environment. Expensive sensors are often left off of consumer grade robotics in order to make them more widely available to consumers. This leaves the robot without effective means of perception. Research has shown that LSTM neural networks can provide insight that would normally only be available to a robot fitted with expensive sensor packages such as RTK GPS or LIDAR systems. Specifically, IMU data paired with controller data and wheel odometry data. By training a LSTM network with the data mentioned from scenarios of a robot that is stuck and a robot that is not stuck a set of networks are experimentally proven to determine if the robot is stuck along, it’s two axes of movement at an accuracy of greater than 70% with a false positive rate less than 12%.
Degree
thesis:*- Name thesis:degree_name
- Master of Science (MS)
- Level thesis:degree_level
- Thesis
- Discipline thesis:degree_discipline
- Mechanical Engineering
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Lysford, Wilson
- Contributors dc:contributor
-
- Jeremiah Neurbert
Subjects
dc:subject × 1Identifiers
dc:identifier.*- Repository record dc:identifier
- https://commons.und.edu/theses/6378