Virtual Sensor System for Sensing Data Estimation of Quadruped Robot
The sensor data measured by the quadruped robot plays a crucial role in recognizing the robot's physical environment and controlling its posture. High accuracy in sensor data is essential for achieving precise robot control. However, most of these high-performance sensors are expensive and often subjected to excessive loads during field operations. This presents significant challenges when considering the practicality of prototypes and the feasibility of mass production, given the associated costs of production, maintenance, and preservation.
To address these issues, this project explores virtual sensor technology as a viable solution. The goal is to reduce dependency on high-cost, high-maintenance sensors by utilizing a virtual sensor system. This system is trained using actual sensor data in combination with a variety of hardware information, thus enabling the substitution of physical sensors with virtual ones. Specifically, this approach employs neural network theory with a focus on back propagation algorithms.
The training process involves the quadruped robot being exposed to different terrains using high-cost sensors to gather accurate data. This data is then fed into a neural network to train the virtual sensors. Once trained, these virtual sensors can effectively replicate the functionality of their physical counterparts, allowing the robot to operate with reduced sensor configurations.
A key achievement of this project is verifying the feasibility of replacing ground reaction force sensors with leg joint encoders on the quadruped robot. This reduction in physical sensors translates to a lower risk of failure, decreased cost, and reduced computational load, all of which are critical factors for field robots.
Future developments in this project aim to expand and refine the virtual sensor scheme to further minimize reliance on costly sensors, thereby enhancing the practicality and scalability of field robot systems. By employing virtual sensor technology, this approach offers a promising path toward more efficient and economically viable robot designs for a variety of applications.
<Virtual Sensor Strategy for Quadruped Robot ‘Jinpoong' developed by KITECH>
- Development of Quadrupdal Dog-horse Platform Technology (KITECH, 2008-2009)
- Development of Quadruped Robot (Capstone Design, 2005-2006)
- Achievements: 1 domestic journal is published, and 6 and 2 papers are presented in domestic and international conferences respectively. Honored ‘Innovative Technology Prize' from a President of Hanyang University (2007)
Virtual Sensor System for Sensing Data Estimation of Quadruped Robot
The sensor data measured by the quadruped robot plays a crucial role in recognizing the robot's physical environment and controlling its posture. High accuracy in sensor data is essential for achieving precise robot control. However, most of these high-performance sensors are expensive and often subjected to excessive loads during field operations. This presents significant challenges when considering the practicality of prototypes and the feasibility of mass production, given the associated costs of production, maintenance, and preservation.
To address these issues, this project explores virtual sensor technology as a viable solution. The goal is to reduce dependency on high-cost, high-maintenance sensors by utilizing a virtual sensor system. This system is trained using actual sensor data in combination with a variety of hardware information, thus enabling the substitution of physical sensors with virtual ones. Specifically, this approach employs neural network theory with a focus on back propagation algorithms.
The training process involves the quadruped robot being exposed to different terrains using high-cost sensors to gather accurate data. This data is then fed into a neural network to train the virtual sensors. Once trained, these virtual sensors can effectively replicate the functionality of their physical counterparts, allowing the robot to operate with reduced sensor configurations.
A key achievement of this project is verifying the feasibility of replacing ground reaction force sensors with leg joint encoders on the quadruped robot. This reduction in physical sensors translates to a lower risk of failure, decreased cost, and reduced computational load, all of which are critical factors for field robots.
Future developments in this project aim to expand and refine the virtual sensor scheme to further minimize reliance on costly sensors, thereby enhancing the practicality and scalability of field robot systems. By employing virtual sensor technology, this approach offers a promising path toward more efficient and economically viable robot designs for a variety of applications.
<Virtual Sensor Strategy for Quadruped Robot ‘Jinpoong' developed by KITECH>
- Development of Quadrupdal Dog-horse Platform Technology (KITECH, 2008-2009)
- Development of Quadruped Robot (Capstone Design, 2005-2006)
- Achievements: 1 domestic journal is published, and 6 and 2 papers are presented in domestic and international conferences respectively. Honored ‘Innovative Technology Prize' from a President of Hanyang University (2007)