cyberpod_sim_ros

The majority of this code was original written by Andrew Singleterry. Adjustments were made by Ryan Cosner and Sarah Dean for use in producing the experimental results for their CoRL 2020 paper: "Guaranteeing Safety of Learned Perception Modulesvia Measurement-Robust Control Barrier Functions" (Arxiv link).

To install and run this package, you'll need a working ROS environment. This repo was tested using Ubuntu 18.04 and Ros Melodic.

Clone the vision packages into your catkin_workspace's src file:

• rviz_camera_stream Rviz plugin that publishes camera streams
• rviz_lighting Rviz plugin that provides ambient lighting

You'll also need the Python ECOS package which can be installed via:

pip install ecos

Running repo

Navigate to your catkin_workspace

cd catkin_workspace

Make the workspace

catkin_make

Source the setup file (this must be done in every new terminal)

source devel/setup.bash

Adjust the epsilon value and launch the main script

roslaunch cyberpod_sim_ros main_mrcbf.launch epsilon:=0.0

Open a new terminal, source the setup file and run the rosservice command to start the simulation:

rosservice call /cyberpod_sim_ros/integrator/ui "cmd: 1 data:-0"

• The commands values are: 0 stops, 1 starts, 4 resets

Recordings

Recordings will appear in the /bags folder in the repo and in that folder is another readme for how to format the data.

Relevant Portions of the Code

• cyberpod_sim_ros/
  • include/
    • dynamics.hpp
  • launch/
    • main_mrcbf.launch
    • mrcbf_with_learning.launch
    • sample_state_space.launch
  • msg/
    • cmd.msg
    • input.mcg
    • learning_data.msg
    • state.msg
  • src/
    • integrator_node.cpp
    • grid_state_space_node.cpp
    • sensor_node.cpp
    • controller_node.cpp
    • src_python/
      • data_logger.py
      • ecos_fitler_node.py
      • perception_node.py
  • srv/
    • ui.srv
  • URDF/
    • cyberpod.urdf

Key Features of Main Simulation:

main_marcbf.launch launches the main simulation

mrcbf_with_learning.launch launches the simulation with learning in the loop (this requires learning data which can be generated using the notebooks). The data couldn't be uploaded due to size. Please contact the Ryan if you would like a copy of the data.

dynamics.hpp the dynamics for the cyberpod as derived from the Euler-Lagrange equations

msg/ the four messages listed above are the custom messages used to relay messages between the nodes of simulation and also between the user and the system

srv/ui.srv the service for the user to interact with the simulation

URDF the system's visual model rviz visualisations

Nodes:

• integrator_node.cpp
  • serves as the main physics engine for the system, receives inputs from ecos_filter_node and outputs integrated states
• perception_node.py
  • recieves image data and outputs state estimates
• sensor_node.cpp
  • receives state values from integrator node state estimates from perception_node and outputs "measured states" with true state values used whenever they cannot be attained from the image, such as velocity.
• controller_node.cpp
  • receives state estimates and output desired input based on PD feedback gains
• mrcbf_filter_node.py
  • receives state estimates and desired input and outputs minimally adjusted input to guarantee safety

Main Code for Data Collection

• sample_state_space.launch
  • launch data collection procedure
• grid_state_space_node.cpp
  • iterates through a gridding of the state space