#social path planner

Social Path Planner

Requirements

Python 3 with the following packages:

• torch
• numpy
• matplotlib
• sklearn
• ROS, project was tested with ROS melodic

note: converting data from spenser tracked msgs to .dense files (as required by network when training) requires python 2 and Spenser tracking msgs installed

Installation

• Checkout or download this repository

• Install ped_sim simulator from https://code.research.uts.edu.au/uts-cas/dwell-track/pedsim_ros/tree/sean/pedsim

  • need to download sean/pedsim version as contains updates to srvs, scenarios and launch files that are necessary.
  • unfortunately as this project actually lives in a different repository and has submodules this bit is a bit hacky. The submodule update as shown in the README on the master branch does not seem to work properly as the sim cannot launch and cannot update project with my own files. A fix is to download sean/pedsim project then do as follows:
    • 'cd 'sean/pedsim-directory'
    • 'cd 2ndparty/spencer_messages' then download https://github.com/spencer-project/spencer_messages/tree/3b392e7e5ba367dd23a3cc07e934e558229437d4 and copy project files into this directory
    • 'cd ../spencer_tracking_rviz_plugin/ then download https://github.com/srl-freiburg/spencer_tracking_rviz_plugin/tree/9433aae7f99cd7b395281ee3af9ff3b629a53b09 and copy the project files into this directory this mimics the submodule update. Then follow the Quick Start Guide

• if you have the annaconda package manager https://docs.conda.io/projects/conda/en/latest/index.html the environment can be initialised by. This was the environment in which I was running the code

  • 'cd social_path_planning'

  • 'conda env create -f social_path_planning.yml'

  • 'conda env list' the social path planning environment should now be listed

  • 'conda activate social_path_planning' to activate environment

Quick Start Guide

open a terminal

• 'cd catkin_ws/src'
• 'ln -s "ped_sim directory"
• 'cd ../'
• 'catkin_make'
• 'roslaunch pedsim_simulator sean.launch'

This should launch the pedestrian simulator as well as the srvs required to get pedestrian data

open new terminal, initialise python environment with the required packages, see above

• 'cd social_path_planning'
• 'source source_me.sh'
• 'cd rnn/apg-disp'
• 'social_planner -w model.plk'

If you wish to train you own model follow instructions at https://code.research.uts.edu.au/10173639/human-motion-rnn. there is a data directory to store data. Run training phase in rnn, it will create a directory refering to model and time of creating.

NOTE functionality to run any model has temporarily been depreciated as experimentation with apg-displacement model is conducted.

Information

As the project uses models from https://code.research.uts.edu.au/10173639/human-motion-rnn and due to the structure of that project there is a lot of interdependency to run the inference of the model. For this reason the developments for this project are living in a very similar environment containing many of the packages from the human-motion-rnn project. The unique contributions of this project are

• social_planner/scripts/social_planner
• social_planner/src/robot.py
• social_planner/src/lstm_motion_model/robot_utils.py
• social_planner/src/lstm_motion_model/robot_plots.py
• social_planner/src/lstm_motion_model/rrt.py

as well as modification to the existing files