Agent Configuration

BiguaSim agents are declared in a list in the scenario dictionary. Each agent is defined using a new dictionary in the list.

Note

The first agent in the agents array is the “main agent”.

"agents":[
   {
      "agent_name": "uav0",
      "agent_type": "{agent types}",
      "sensors": [
         "array of sensor objects"
      ],
      "location": [1.0, 2.0, 3.0],
      "rotation": [1.0, 2.0, 3.0],
      "location_randomization": [1, 2, 3],
      "rotation_randomization": [10, 10, 10],
      "control_abstraction": "{control abstraction type}"
      "dynamics" : {
             "batch_size" : 1,
      },
   },
   {
      "agent_name": "uav1"
      ...
   }
]

Below is a description of the keys in the agent dictionary:

Agent Name

agent_name is a string that specifies the name of the agent. This name is used to identify the agent when interacting with the environment.

Agent Type

agent_type is a string that specifies the type of agent. This table gives the current valid arguments for agent_type:

Agent Type	String in agent_type
BlueBoat	`BlueBoat`
TorpedoAUV	`TorpedoAUV`
BlueROV2	`BlueROV2`
BlueROV2 Heavy	`BlueROV2 Heavy`
DJI Matrice	`DjiMatrice`

Details on each agent can be found at the individual agent pages in BiguaSim Agents. This list will be updated as more agents are added to BiguaSim.

Sensors

sensors is an array of sensor objects that are attached to the agent. Each sensor is defined using a new dictionary in the list. For details on configuring and using sensors, see Sensor Configuration.

Location and Rotation

These keys define the location and orientation of the agent in the world, measured in meters and degrees respectively. The location is in the format [dx, dy, dz] and the rotation is [roll, pitch, yaw], rotated about XYZ fixed axes, ie R_z R_y R_x.

Note

Location uses BiguaSim world coordinates, not Unreal Engine level coordinates!
BiguaSim coordinates are right handed in meters.
See Units and Coordinates in BiguaSim for details.

Location Randomization

location_randomization and rotation_randomization are optional. If provided, the agent’s start location and/or rotation will vary by a random amount sampled uniformly from the specified range. Randomization in each direction is sampled independently.

The location randomization value is measured in meters, in the format [dx, dy, dz]. The rotation randomization is in the format [roll, pitch, yaw], rotated about the XYZ fixed axes (i.e. R_z*R_y*R_x).

Control Abstraction

The control_abstraction string defines how the simulator interprets commands sent to an agent via the step or tick methods. It acts as the interface layer between high-level autonomous logic and low-level physics.

Standardized Agents

In BiguaSim, most agents (such as the BlueROV2 and HoveringAUV) share a unified set of control abstractions. These allow for seamless switching between different levels of autonomy:

``cmd_vel``: Controls linear velocity $(v_x, v_y, v_z)$ .
``cmd_vel_yaw``: Similar to cmd_vel, but maintains a specific target yaw heading during maneuvers.
``cmd_pos_yaw``: A high-level abstraction where the user provides target coordinates $[x, y, z]$ and a target yaw. An internal PID controller handles the trajectory.
``accel``: Provides direct acceleration commands along the agent’s axes.
``cmd_motor_speeds``: The lowest level of control for standard agents, allowing users to send individual thrust values to each motor (e.g., 8 independent thrusters for the BlueROV2).

The TorpedoAUV Exception

The TorpedoAUV does not follow the standard multi-thruster abstractions due to its unique underactuated physical design (single rear propeller and control surfaces). It uses a specialized abstraction:

``cmd_rudders_sterns_motor_speed``: Requires a specific vector to control the main motor RPM and the deflection angles of the rudders and stern planes.
- Input Vector: [Motor_RPM, Rudder1, Rudder2, Stern1, Stern2]

For more details on specific control implementation and PID tuning, see Control Abstractions.