salon_ws

Open Source Implementation of SALON

Documentation coming soon, but this repo is set up as a ROS1 workspace. The launch files in src/context_adaptation will run the code necessary to reproduce the examples on different robots, specifically:

• online_hdif.launch
• online_hdif_wheelchair.launch
• online_hdif_anymal.launch

---

🚀 Getting Started

The main requirements are ros 1 (or ros 2 but these examples are for ros 1), and pytorch. The rest of the packages should be easy to install.

1. Clone the workspace

git clone https://github.com/castacks/salon_ws.git
cd salon_ws
catkin build
source devel/setup.bash

2. Download Data

Off-road ATV

Use the TartanDrive 2.0 GUI to download a sample bag. A good higher-speed example is "2023-11-14-14-26-22_gupta". The corresponding launch file is in src/context_adaptation/online_hdif.launch

Wheelchair

Data will be available soon. The corresponding launch file is in src/context_adaptation/online_hdif_wheelchair.launch

Wild Visual Navigation

Follow the instructions to set up Wild Visual Navigation separately, as we use their velocity-tracking cost function that the provide. As mentioned here it needs to be tuned a bit to get something more expressive on their sample data. A good example to run on is their "MPI_Outdoor_Seq_1" example.

The corresponding launch file is in src/context_adaptation/online_hdif_anymal.launch

---

🧪 Running the System

Recommended setup is as follows

Terminal 1:

roscore

Terminal 2:

roslaunch context_adaptation online_hdif.launch

Replace online_hdif.launch with the appropriate launch file

Terminal 3:

rosbag play **bag_name**

Terminal 4: Open up rviz and observe the "shortrange_costmap" and "shortrange_speedmap" rgb_viz layers.

If the node doesn't start producing maps soon after the bag starts, try using sim time (i.e. rosparam set use_sim_time true and use --clock in rosbag play)

---

If you use our work please consider citing:

     @misc{sivaprakasam2024salonselfsupervisedadaptivelearning,
        title={SALON: Self-supervised Adaptive Learning for Off-road Navigation}, 
        author={Matthew Sivaprakasam and Samuel Triest and Cherie Ho and Shubhra Aich and Jeric Lew and Isaiah Adu and Wenshan Wang and Sebastian Scherer},
        year={2024},
        eprint={2412.07826},
        archivePrefix={arXiv},
        primaryClass={cs.RO},
        url={https://arxiv.org/abs/2412.07826}, 
  }