Adding a New Robot to RoboEval

Adding a New Robot to RoboEval#

To integrate a new robot into RoboEval, you’ll need to define its configuration and register it within the system. This guide walks through the necessary steps using the Franka Panda robot as an example.

Step 1: Create a Robot Configuration File#

Navigate to:

roboeval/robots/configs/

Create a new Python file (e.g., panda.py) where you define the robot’s configuration.

Required Components#

Each robot should include:

  • ArmConfig – MuJoCo model path, joint and actuator definitions

  • GripperConfig – model path, control range, and finger pad bodies

  • FloatingBaseConfig – defines fixed or floating base behavior

  • CameraConfig – (optional) cameras on base or wrists

  • RobotConfig – wraps all robot components

  • RobotIKConfig – IK settings used by CuRobo or similar planners

  • Robot subclass – inherits from roboeval.robots.robot.Robot and returns config and ik_config

Example: Define the Robot Class#

class BimanualPanda(Robot):
    @property
    def config(self) -> RobotConfig:
        return PANDA_CONFIG_WITH_PANDA_GRIPPER

    @property
    def ik_config(self) -> RobotIKConfig:
        return create_bimanual_panda_config()

Example: ArmConfig Definition#

PANDA_ARM = ArmConfig(
    site="attachment_site",
    model=...,
    joints=[...],
    actuators=[...],
    links=[...],
)

Step 2: Register the Robot#

To make the robot available for use in RoboEval experiments, register it in:

tools/shared/utils.py

Add your robot class to the ROBOTS dictionary:

from roboeval.robots.configs.panda import BimanualPanda, SinglePanda

ROBOTS: dict[str, Optional[Type[Robot]]] = {
    "Default": None,
    "Bimanual Panda": BimanualPanda,
    "Single Panda": SinglePanda,
    # Add your custom robots here
}

Note

The dictionary key (e.g., "Bimanual Panda") is the name used in config files or CLI commands.

Step 3: Tips and Best Practices#

  • Modular Configs: Use shared variables (e.g., PANDA_ARM) for reusable parts across multiple robot configs.

  • Camera Support: Use CAMERA_CONFIG to define wrist and head cameras. Set manual=True and specify pose manually.

  • Floating Base: Use FloatingBaseConfig if the robot has a movable base (e.g., mobile or virtual pelvis).

  • Joint Limits: Ensure joint limits in RobotIKConfig are precise to enable proper inverse kinematics.

Final Checklist#

Task

Done?

Added ArmConfig, GripperConfig, FloatingBaseConfig, RobotConfig

[ ]

Defined RobotIKConfig for motion planning

[ ]

Created a subclass of Robot with config and ik_config properties

[ ]

Registered the robot in tools/shared/utils.py

[ ]

Once registered, your robot is fully compatible with RoboEval, and can be used for benchmarking, evaluation, and policy testing.

Contents