Trajectory tracking based on multi-layered continuous network
Simultaneous Localization and Mapping is widely recognized as a fundamental component in the realm of robotics. Over the decades, researchers are trying to combine the neuron studies in the brain of mammalian into the SLAM system. Attractor network, which is the current model of grid cells is drawin...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/175625 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Simultaneous Localization and Mapping is widely recognized as a fundamental component in the realm of robotics. Over the decades, researchers are trying to combine the neuron studies in the brain of mammalian into the SLAM system. Attractor network, which is the current model of grid cells is drawing attention since it can provide a representation of possibility distribution of the current states. This representation can not only be decoded for the current position, but also naturally suit for fusion perception.
However, simulating attractor network could be tough in computers: the number of cells is quadratic to the size of network, and the impact link is quadratic to the number of cells. In addition, the detail process of attractor network haven't been conclusive.
In this thesis, we formulated the model of attractor neuron dynamic into controversial calculation, making the large scale simulation possible. We also compressed the calculation complexity by introducing the multilayered continues network and cross layer update procedure to further extend usability to vary large scale situation. A boundary cross detection and handle protocol is introduced to reserve the wrapped natural of attractor network. A series of experiments and tests are performed on Kitti dataset alongside with trajectories generated from the City Scale Navigation Simulation. Result shows ATE of 4.46m and marginal ATE increase rate 8.354 m/km on Kitti datasets |
|---|