FlightGoggles : Multiple agents across multiple machines

In this scenario, we consider the problem of simulation of multiple agents using multiple host computers on a local network. For connecting computers that are not on the same network, please consider Setting up a VPN Server and Client.

In order to use multiple machines, FlightGoggles client in each machine needs to be started with different input and output ports. Also, the IP address of the Python client has to be passed as the argument.

These are the example script to execute multiple renderers in different machines:

./FlightGogglesv3.x86_64 -client-ip "client.ip.address" -input-port 10253 -output-port 10254
./FlightGogglesv3.x86_64 -client-ip "client.ip.address" -input-port 10255 -output-port 10256

These ports configurations have to be added in the renderer block of FlightGogglesClients.yaml.

After registering renderers, the user can configure which renderer that each camera will use in the camera_model block. It is possible to simulate a stereo camera by attaching two cameras into one vehicle model with different relative poses. The unattached camera will not update its position and record the video in the initial position.

state:
    sceneFilename: "Stata_GroundFloor"
    camWidth: 640
    camHeight: 480
    camFOV: 70.0
    camDepthScale: 0.20

renderer:
    0:
        inputPort: "10253"
        outputPort: "10254"
    1:
        inputPort: "10255"
        outputPort: "10256"

objects:
    0:
         ID: uav1
         prefabID: Blackeagle
         size_x: 5
         size_y: 5
         size_z: 5

camera_model:
    0:
        ID: cam0
        channels: 3
        renderer: 0
        freq: 30
        outputShaderType: -1
        hasCollisionCheck: False
    1:
        ID: cam1
        channels: 3
        renderer: 1
        freq: 30
        outputShaderType: -1
        hasCollisionCheck: False
    2:
        ID: cam2
        channels: 3
        renderer: 1
        freq: 30
        outputShaderType: -1
        hasCollisionCheck: False
        initialPose: [-10.5, -18.5, -2, 1.0, 0, 0, 0]

vehicle_model:
    uav1:
        type: "uav"
        initialPose: [-6.5, -18.5, -2,  0.707, 0, 0, -0.707]
        imu_freq: 200
        objectsInfo:
            uav1:
                relativePose: [0, 0, 0.3, 0.707, 0, 0, -0.707]
        cameraInfo:
            cam0:
                relativePose: [0.2, 0, 0, 1, 0, 0, 0]
            cam1:
                relativePose: [-0.2, 0, 0, 1, 0, 0, 0]

This is the sample code to test the multiple machine configuration:

import numpy as np
from flightgoggles.env import flightgoggles_env

if __name__ == "__main__":
    env = flightgoggles_env()
    for i in range(200):
        env.proceed_motor_speed("uav1", np.ones(4)*1133.0, 0.01)
    
    ani_set = env.plot_state_video()
    if "cam0" in ani_set.keys():
        display(HTML(ani_set["cam0"].to_html5_video()))
    if "cam1" in ani_set.keys():
        display(HTML(ani_set["cam1"].to_html5_video()))
    if "cam2" in ani_set.keys():
        display(HTML(ani_set["cam2"].to_html5_video()))
    env.close()

Fig 1. The result of the example code (cam 0)

Fig 2. The result of the example code (cam 1)

Fig 3. The result of the example code (cam 2)

FlightGoggles : Multiple agents across multiple machines

Attachments: