Software Overview

ROBOTIS Hand is a dexterous robotic hand system designed for real-world Physical AI research and manipulation tasks. It supports high-DOF finger control and enables smooth integration of both teleoperation and AI policy execution.

The platform runs on ROS 2 Jazzy and uses the ros2_control framework for real-time joint-level control. Each hand features 20 degrees of freedom powered by XM335 actuators operating in position mode via TTL communication. The actuators are connected through a Hand Controller Board that converts RS-485 signals from the U2D2 interface to 5 TTL channels, enabling distributed control of finger joints. The system also integrates tactile sensors that provide feedback through ros2_control state interfaces, enabling rich sensory perception for manipulation tasks.

This platform is designed for:

Collecting motion data through teleoperation
Training reinforcement learning or imitation learning models
Running AI-generated trajectories in real time without changing system code

It is suitable for researchers, developers, and integrators working with AI-enabled robotics.

System Architecture

The diagram below illustrates the overall control structure of ROBOTIS Hand.
External teleoperation or trajectory commands are received via ROS 2 topics, processed in real time by ros2_control, and executed by DYNAMIXEL Actuators via TTL.

software_architecture

Layer	Component	Description
Motion Control	`ros2_control`	100Hz joint control loop
Actuators	XM335	Position mode via TTL
Hand Controller Board	PCB Board (RS-485)	RS-485 to 5 TTL channels
Communication	U2D2 (RS‑485)	4 Mbps, Dynamixel Protocol 2.0
Sensors	Tactile sensors	Read via ros2_control state_interfaces

Why `ros2_control`?

ros2_control is a real-time, modular control framework used in ROS 2. ROBOTIS Hand uses it without major changes.

Separates control logic from hardware drivers
Operates at a fixed 100Hz update rate
Supports dynamic loading or replacement of controllers
Allows AI policies to send trajectory commands using standard ROS topics

This structure supports both manual operation and AI-based control in the same system.

Motion Execution Pipeline

Input Source (Teleoperation)
      ↓
ROS 2 JointTrajectory Topic
      ↓
controller_manager (100Hz)
      ↓
JointTrajectoryController(s)
      ↓
resource_manager (interface arbitration)
      ↓
DynamixelHardwareInterface (position / current)
      ↓
RS‑485 via Dynamixel SDK
      ↓
DYNAMIXEL Actuators

Step-by-Step

Trajectory generation — Created by VR, GUI, or AI model
ROS topic publishing — Commands sent to /leader/joint_trajectory_command_broadcaster_*
Controller manager — Runs the main read → update → write loop at 100Hz
JointTrajectoryController — Splits trajectory and commands each joint
Resource manager — Controls access to command interfaces and prevents conflicts between controllers
Hardware interface — Converts commands to RS-485 packets using Dynamixel SDK
Actuators — Execute the motion and return position/current feedback

Controller Configuration & Joint Mapping

Controller	Segment	DOF	Input Topic
`hand_l_controller`	Left hand	20	`/leader/joint_trajectory_command_broadcaster_left_hand/joint_trajectory`
`hand_r_controller`	Right hand	20	`/leader/joint_trajectory_command_broadcaster_right_hand/joint_trajectory`
`joint_state_broadcaster`	All joints	–	Publishes to `/joint_states`

All controllers (except joint_state_broadcaster) use the JointTrajectoryController type.

By default, all joints operate in position mode.

Controller YAML Example (link)

yaml

/**:
  controller_manager:
    ros__parameters:
      use_sim_time: False
      update_rate: 100  # Hz

      joint_state_broadcaster:
        type: joint_state_broadcaster/JointStateBroadcaster

      left_hand_controller:
        type: joint_trajectory_controller/JointTrajectoryController

      right_hand_controller:
        type: joint_trajectory_controller/JointTrajectoryController

      left_effort_controller:
        type: effort_controllers/JointGroupEffortController

      right_effort_controller:
        type: effort_controllers/JointGroupEffortController

/**:
  left_hand_controller:
    ros__parameters:
      joints:
        - finger_l_joint1
        - finger_l_joint2
        - finger_l_joint3
        - finger_l_joint4
        - finger_l_joint5
        - finger_l_joint6
        - finger_l_joint7
        - finger_l_joint8
        - finger_l_joint9
        - finger_l_joint10
        - finger_l_joint11
        - finger_l_joint12
        - finger_l_joint13
        - finger_l_joint14
        - finger_l_joint15
        - finger_l_joint16
        - finger_l_joint17
        - finger_l_joint18
        - finger_l_joint19
        - finger_l_joint20
      command_interfaces:
        - position
      state_interfaces:
        - position
        - velocity
      allow_nonzero_velocity_at_trajectory_end: true
      allow_partial_joints_goal: true

/**:
  right_hand_controller:
    ros__parameters:
      joints:
        - finger_r_joint1
        - finger_r_joint2
        - finger_r_joint3
        - finger_r_joint4
        - finger_r_joint5
        - finger_r_joint6
        - finger_r_joint7
        - finger_r_joint8
        - finger_r_joint9
        - finger_r_joint10
        - finger_r_joint11
        - finger_r_joint12
        - finger_r_joint13
        - finger_r_joint14
        - finger_r_joint15
        - finger_r_joint16
        - finger_r_joint17
        - finger_r_joint18
        - finger_r_joint19
        - finger_r_joint20
      command_interfaces:
        - position
      state_interfaces:
        - position
        - velocity
      allow_nonzero_velocity_at_trajectory_end: true
      allow_partial_joints_goal: true

/**:
  left_effort_controller:
    ros__parameters:
      joints:
        - finger_l_joint1
        - finger_l_joint2
        - finger_l_joint3
        - finger_l_joint4
        - finger_l_joint5
        - finger_l_joint6
        - finger_l_joint7
        - finger_l_joint8
        - finger_l_joint9
        - finger_l_joint10
        - finger_l_joint11
        - finger_l_joint12
        - finger_l_joint13
        - finger_l_joint14
        - finger_l_joint15
        - finger_l_joint16
        - finger_l_joint17
        - finger_l_joint18
        - finger_l_joint19
        - finger_l_joint20

/**:
  right_effort_controller:
    ros__parameters:
      joints:
        - finger_r_joint1
        - finger_r_joint2
        - finger_r_joint3
        - finger_r_joint4
        - finger_r_joint5
        - finger_r_joint6
        - finger_r_joint7
        - finger_r_joint8
        - finger_r_joint9
        - finger_r_joint10
        - finger_r_joint11
        - finger_r_joint12
        - finger_r_joint13
        - finger_r_joint14
        - finger_r_joint15
        - finger_r_joint16
        - finger_r_joint17
        - finger_r_joint18
        - finger_r_joint19
        - finger_r_joint20

Debugging & Visualization Tools

Tool / Topic	Description
`ros2 control list_controllers`	List loaded controllers and their status
`/joint_states`	Real-time joint position and velocity feedback
RViz2	3D view of robot model (URDF), TF, and movement

Safety & Fault Handling

Joint limits are enforced using the URDF and controller configuration
Out-of-range values are clamped by the hardware interface (YAML-defined limits)
The Dynamixel hardware interface checks for communication errors with each motor

Software Overview ​

System Architecture ​

Why ros2_control? ​

Motion Execution Pipeline ​

Step-by-Step ​

Controller Configuration & Joint Mapping ​

Controller YAML Example (link) ​

Debugging & Visualization Tools ​

Safety & Fault Handling ​