Skip to content

Latest commit

 

History

History
97 lines (87 loc) · 4.45 KB

README.md

File metadata and controls

97 lines (87 loc) · 4.45 KB
Raw

SOGMP++/SOGMP: Stochastic Occupancy Grid Map Prediction in Dynamic Scenes

Notice

The new version with software optimization is available at "SCOPE: Stochastic Cartographic Occupancy Prediction Engine for Uncertainty-Aware Dynamic Navigation", where we used the acronym SCOPE (Stochastic Cartographic Occupancy Prediction Engine) instead of SOGMP. The updated code is available in our new GitHub repository scope and its related predictive uncertainty-aware navigation code is available in our new GitHub repository scope_nav.

Introduction

Implementation code for our paper "Stochastic Occupancy Grid Map Prediction in Dynamic Scenes"(arXiv) in Conference on Robot Learning (CoRL) 2023. Two stochastic occupancy grid map (OGM) predictor algorithms (i.e. SOGMP and SOGMP++) were implemented by Pytorch. Video demos can be found at multimedia demonstrations. Here are three GIFs showing the occupancy grid map prediction comparison results (0.5s, or 5 time steps into the future) of our proposed SOGMP++, SOGMP algorithms, and ConvLSTM, PhyDNet, DeepTracking, and SOGMP_NEMC baselines on three different datasets with different robot models.

  • OGM-Turtlebot2: turtlebot2_prediction_demo
  • OGM-Jackal: jackal_prediction_demo
  • OGM-Spot: spot_prediction_demo

Requirements

  • python 3.7
  • torch 1.7.1
  • tensorboard

OGM-Datasets

The related datasets can be found at OGM-datasets
There are three different datasets collected by three different robot models (i.e. Turtlebot2, Jackal, Spot).

  • 1.OGM-Turtlebot2: collected by a simulated Turtlebot2 with a maximum speed of 0.8 m/s navigates around a lobby Gazebo environment with 34 moving pedestrians using random start points and goal points
  • 2.OGM-Jackal: extracted from two sub-datasets of the socially compliant navigation dataset (SCAND), which was collected by the Jackal robot with a maximum speed of 2.0 m/s at the outdoor environment of the UT Austin
  • 3.OGM-Spot: extracted from two sub-datasets of the socially compliant navigation dataset (SCAND), which was collected by the Spot robot with a maximum speed of 1.6 m/s at the Union Building of the UT Austin

Usage: SOGMP (The inference speed is faster than SOGMP++)

  • Download OGM-datasets and decompress them to the home directory:
cd ~
tar -zvxf OGM-datasets.tar.gz
  • Training:
git clone https://github.com/TempleRAIL/SOGMP.git
cd SOGMP 
git checkout sogmp
sh run_train.sh ~/data/OGM-datasets/OGM-Turtlebot2/train ~/data/OGM-datasets/OGM-Turtlebot2/val
  • Inference Demo on OGM-Turtlebot2 dataset:
git clone https://github.com/TempleRAIL/SOGMP.git
cd SOGMP 
git checkout sogmp
sh run_eval_demo.sh  ~/data/OGM-datasets/OGM-Turtlebot2/test

Usage: SOGMP++ (The prediction accuracy is higher than SOGMP)

  • Download OGM-datasets and decompress them to the home directory:
cd ~
tar -zvxf OGM-datasets.tar.gz
  • Training:
git clone https://github.com/TempleRAIL/SOGMP.git
cd SOGMP 
git checkout sogmp++
sh run_train.sh ~/data/OGM-datasets/OGM-Turtlebot2/train ~/data/OGM-datasets/OGM-Turtlebot2/val
  • Inference Demo on OGM-Turtlebot2 dataset:
git clone https://github.com/TempleRAIL/SOGMP.git
cd SOGMP 
git checkout sogmp++
sh run_eval_demo.sh  ~/data/OGM-datasets/OGM-Turtlebot2/test

Citation

@inproceedings{xie2023sogmp,
  doi = {10.48550/ARXIV.2210.08577},
  title={Stochastic Occupancy Grid Map Prediction in Dynamic Scenes},
  author={Zhanteng Xie and Philip Dames},
  booktitle={Proceedings of The 7th Conference on Robot Learning},
  pages={1686--1705},
  year={2023},
  volume={229},
  series={Proceedings of Machine Learning Research},
  month={06--09 Nov},
  publisher={PMLR},
  url={https://proceedings.mlr.press/v229/xie23a.html}
}


@article{xie2023stochastic,
  title={Stochastic Occupancy Grid Map Prediction in Dynamic Scenes},
  author={Xie, Zhanteng and Dames, Philip},
  journal={arXiv preprint arXiv:2210.08577},
  year={2023}
}