MISC

Deep Visual Odometry

For this project, I implemented DeepVO based on arxiv1 and arxiv2, a deep visual odometry model, to estimate the motion of a robot from a sequence of camera images.

Period: 2020.03 - 2020.07

code

Our team implemented DeepVO and integrated it with the ROS Navigation stack. To test its effectiveness, we added Gaussian noise to the wheel odometry data and compared the navigation performance with and without DeepVO in the localization algorithm. Ultimately, we added DeepVO as a measurement model to the EKF (Extended Kalman Filter) localization algorithm.

Image 1
Robot Hardware Configuration.
 Image 2
Navigation Software Architecture.


Navigation Experiment Environment. (The red dot is the target destination.)
Image 1
Environment Scene 1.
 Image 2
Environment Scene 2.


Navigation Experiment Result. (The red dot is the target destination.)
Image 1
Navigation without DeepVO.
 Image 2
Navigation with DeepVO.


MVAE Multi-Sensor Data Fusion for Robotic Arms

For my graduation thesis, I conducted a study on fusing robot sensor data using an MVAE model.

Period: 2020.03 - 2020.06

Image 1
Simulation Env for Robotic Data Collection.
 Image 2
An Example of Collected Data.


Image 1
MVAE Neural Network Architecture.
 Image 2
Data generation by manipulating the values of each dimension of the latent vector.


Autonomous AED Delivery Robot

For my graduation project, I designed and built an autonomous delivery robot from the ground up, including all the hardware and software components.

Period: 2019.09 - 2019.12

Image 1
3D CAD Modeling.
 Image 2
Final Product.


Overall System Architecture.


Scenario.