IMOST: Incremental Memory Mechanism with Online Self-Supervision for Continual Traversability Learning

Kehui Ma  Zhen Sun  Chaoran Xiong  Qiumin Zhu  Kewei Wang  Ling Pei^* 

Shanghai Jiao Tong University

^*Corresponding author

Based on the wild_visual_navigation, we have expanded the online continual learning task for traversability. And we have deployed the algorithm on Unitree Go2 quadruped robot which is equipped with a LIVOX MID-360 LiDAR and a depth camera D435i. We use FastLIO to provide localization information. All the code is deployed in Docker on the Jetson AGX Orin platform.

Docker Environment Setup

The Docker development environment supports ARM and AMD platforms separately. The ARM Docker image is built on jetpack_r35.2.1, while the AMD Docker image is built on pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime. The detailed information is showed in dockerfile.

Only for Test

Todo

Complete version

Todo

Citation

If you find this project helpful, please consider citing us:

@article{ma2024imost,
  doi = {10.13140/RG.2.2.33195.86560},
  url = {https://rgdoi.net/10.13140/RG.2.2.33195.86560},
  author = {{Kehui Ma} and Sun, Zhen and {Chaoran Xiong} and {Qiumin Zhu} and {Kewei Wang} and Pei, Ling},
  language = {en},
  title = {IMOST: Incremental Memory Mechanism with Online Self-Supervision for Continual Traversability Learning},
  publisher = {Unpublished},
  year = {2024}
}