Overview

This repo is intended to be used as part of evaluation 1 for the DARPA ANSR project. NOTE: We are a manuever thread performer for evaluation 1.

Requirements

This repo requires a Ubuntu 22.04 Linux environment capable of running the ADK Airsim docker images system with an nvidia GPU with Cuda and docker.

The minimum required VM size is: NC12s_v3 with a 12 core Intel Xeon E5-2690 v4 CPU with 224 GB RAM and dual Nvidia V100s with a total VRAM of 32 GB. Note: we actually run our docker container on a machine with a 16 core 13th Gen Intel® Core™ i9-13900K CPU with 32 GB RAM and a RTX4090 with VRAM of 24GB.

We tested our code with ADK version 2.1.5 and we expect our code to be evaluated using this ADK image. We intend to use flight altitude (MAP_ALTITUDE) of 15m.

Testing

Our image can be tested by using the deploy docker compose file. Our docker container contains all the neccessary mounted volumes and only relies on the packages built into the container.

docker compose -f docker-compose.deploy.yml up

A sample deploy docker compose file is as follows:

version: "3.8"
# name: ansr-collins
services:
  adk:
    image: darpaansr.azurecr.io/adk:2.1.5
    volumes:
      - "/tmp/.X11-unix:/tmp/.X11-unix:rw"
      - "./adk/mission_briefing:/mission_briefing"
      - "./adk/output:/output"
    command: "--mission_thread=manuever_thread --mission_class=area_search --restart" # Comment when using custom mission config
    ipc: host
    pid: host
    environment:
      - DISPLAY
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
  deployment:
    image: darpaansr.azurecr.io/collins:maneuver-1.0.5
    ipc: host
    pid: host
    volumes:
      - "./adk/mission_briefing:/mission_briefing"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [ gpu ]

Our code is automatically executed when the docker container is deployed and we achieve this using the ros_node.sh script with runs the following commands in the deployment container:

source "/home/performer/dev_ws/install/setup.bash"
python3 "/home/performer/dev_ws/src/verifiable-compositional-rl/src/ansr_eval1/waypoint_publisher.py"

Since we are a manuever thread performer for evaluation 1, you may echo the adk_node/input/sparse_waypoints topic to view the waypoints we publish.

docker compose exec deployment bash
(in the container) source install/setup.bash
(in the container) ros2 topic echo adk_node/input/waypoints

To run in headless mode use the following compose file instead:

Run the following in one terminal:

cd ta3-documentation
sh start_adk_sparse_manuever.sh --mission_thread=manuever_thread -f 20

In another terminal run the following compose file using: docker compose -f docker-compose.deploy.yml up:

version: "3.8"
# name: ansr-collins
services:
  deployment:
    image: darpaansr.azurecr.io/collins:maneuver-1.0.5
    ipc: host
    pid: host
    volumes:
      - "./adk/mission_briefing:/mission_briefing"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [ gpu ]

Building

Run the following in one terminal to build the container with local changes:

docker build -t ansr-collins .

High-level overview of our code (inputs/outputs)

Our main algorithm is housed inside the file named waypoint_publisher.py at /home/performer/dev_ws/src/verifiable-compositional-rl/src/ansr_eval1/waypoint_publisher.py.

We have an rclpy node instance named mission_exec and where we house a ros node named waypoint_publisher.

We subscribe to the following topics:

adk_node/ground_truth/perception adk_node/SimpleFlight/odom_local_ned

and we publish the following topics:

adk_node/input/sparse_waypoints adk_node/input/terminate

We have commented our code for further explanation. Our code cannot be run without ROS at present.