Thesis
ENABLING PRECISE CONTROL OF A HAPTIC DEVICE: A MACHINE LEARNING APPROACH
Washington State University
Master of Science (MS), Washington State University
01/2021
DOI:
https://doi.org/10.7273/000001858
Handle:
https://hdl.handle.net/2376/119872
Abstract
Magnetorheological brakes (MRB) with electronic control can be used in haptic devices to apply forces/torques to the user in a virtual reality (VR) simulation to increase realism. With a control device that uses a Hall sensor to calculate the magnetic field, precise control of the braking torque is possible. Many scientific studies in recent years have shown that machine learning models can be used to predict output torque using Hall sensor data. However, fluid leaks out of the MRB over time due to rubber seal failure, degrading haptic interface output and posing challenges in torque prediction. The negative impacts of fluid leaks on machine learning-based torque prediction have not been well studied, and there is no prior work on alleviating these negative impacts.In this thesis, I address this challenge by developing various machine learning-based methods for capturing the dynamic behavior of an MRB and its changing torque production as fluid leaks out. These approaches include the random forest model, which is based on decision trees, the artificial neural network model, which is based on Multi-Layer Perceptron, and the Long Short-Term Memory model, which is based on Recurrent Neural Network. Extensive experiments using data obtained from a real MRB device have been conducted, and the results show that the both the traditional machine learning approach (2-Step-RN) and the deep learning based approach (Long Short-Term Memory, i.e., LSTM) can reliably predict the output torque. It is worth noting that these approaches outperform baseline models that are trained for and work at a stable fluid level, suggesting their great potential for allowing high-fidelity torque control of MRB devices. In addition, due to various hardware limitations that may be faced, how to combine models with hardware devices is also a challenge. Hence, a lookup table solution and LSTM with interpolation method are proposed in this thesis to address the hardware challenges.
Metrics
3 File views/ downloads
29 Record Views
Details
- Title
- ENABLING PRECISE CONTROL OF A HAPTIC DEVICE: A MACHINE LEARNING APPROACH
- Creators
- Xinyu Zheng
- Contributors
- Xinghui Zhao (Advisor)Xinghui Zhao (Committee Member)Hakan Gurocak (Committee Member)Ben McCamish (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- Engineering and Computer Science (VANC), School of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University
- Number of pages
- 62
- Identifiers
- 99900606550301842
- Language
- English
- Resource Type
- Thesis