Simulation
The Simulation environment utilizes Gazebo and Aerial Gym Simulator. Gazebo is used to test and validate the complete autonomy stack in realistic 3D environments, while Aerial Gym Simulator provides a massively parallelized environment for training data-driven and learning-based methods.
Gazebo
We provide ROS 2 simulators for multi-rotor and differntial drive wheeled robot integrated with the Unified Autonomy Stack. Both simulators are located in the unified_autonomy_stack/workspaces/ws_sim workspace.
Multirotor Simulator
The multirotor simulator consists of the ROS 2 package unified_autonomy_stack/workspaces/ws_sim/src/rmf_gz.
The simulator provides two robots each carrying a color camera, a depth camera, and a LiDAR sensor. The difference is in the type of lidar onboard the robot.
For both robots, the following sensors and control options are setup. <ns> referes to the namespace for that robot.
Sensors:
| Sensor | Topic | Data type | Rate (Hz) |
|---|---|---|---|
| Color camera | /<ns>/cam/rgb |
sensor_msgs/msg/Image |
10 |
| Depth camera (Image) | /<ns>/cam/depth |
sensor_msgs/msg/Image |
30 |
| Depth camera (Point Cloud) | /<ns>/cam/pc |
sensor_msgs/msg/PointCloud2 |
30 |
| LiDAR | /<ns>/lidar/points |
sensor_msgs/msg/PointCloud2 |
10 |
Control Options
| Command Type | Topic | Data type |
|---|---|---|
| Velocity | /<ns>/cmd/vel |
geometry_msgs/msg/Twist |
| Acceleration | /<ns>/cmd/vel |
geometry_msgs/msg/Twist |
Sensor configuration 1: Traditional LiDAR
Robot namespace (<ns>): rmf
LiDAR configuration:
- Field of View:
360 x 90 deg - Orientation:
Sensor configuration 2: Dome LiDAR
Robot namespace (<ns>): rmf_unipilot
LiDAR configuration:
- Field of View: Dome
360 x 90 deg - Orientation:
Wheeled Robot Simulator
The wheeled robot simulator consists of the ROS 2 package unified_autonomy_stack/workspaces/ws_sim/src/ugv_gz.
The simulator provides a differential drive wheeled robot carrying a color camera, a depth camera, and a LiDAR sensor.
The following sensors and control options are setup.
Sensors:
| Sensor | Topic | Data type | Rate (Hz) |
|---|---|---|---|
| Color camera | smb_arl/cam/rgb |
sensor_msgs/msg/Image |
10 |
| Depth camera (Image) | smb_arl/cam/depth |
sensor_msgs/msg/Image |
30 |
| Depth camera (Point Cloud) | smb_arl/cam/pc |
sensor_msgs/msg/PointCloud2 |
30 |
| LiDAR | smb_arl/lidar/points |
sensor_msgs/msg/PointCloud2 |
10 |
Control Options
| Command Type | Topic | Data type |
|---|---|---|
| Velocity | smb_arl/cmd/vel |
geometry_msgs/msg/Twist |
Aerial Gym Simulator
The Aerial Gym Simulator is an open-source simulator for massively parallelized simulation of multirotor platforms.
To train policies for the autonomy stack, a dedicated training task description is set up within the simulator and can be found [here]. This includes the environment setup, reward formulation, obstacle management and episode management. The training task uses a simulated quadrotor platform with a dome LiDAR sensor placed, facing backwards.