Ned3 Pro Adaptive Controller (No Torque Interface)

Role

Developer - Robotics Control Systems

Project Summary

I built a custom controller for the Ned3 Pro focused on one core problem: running an adaptive control law when the practical motor interface does not provide a direct torque command path.

The result was a working ROS1 controller that executes the control law in torque-like form, then safely bridges it to current-mode actuation on hardware.

Core Challenge

• The control design is expressed in torque-domain form.
• The practical motor interface on Ned3 Pro had to be driven through current-mode commands (Mode 0), not direct torque writes.
• This required a reliable torque-to-current bridge with runtime safety constraints.

What I Built

• A custom ros_control plugin:
  • niryo_robot_paper_controller/NeuralAdaptivePidController
• Hardware integration for Ned3 Pro joint control in a private single-joint configuration.
• Real-time command bridge:
  • computes adaptive controller output in torque-like form
  • maps output into safe current-domain actuation
  • sends low-level motor commands through the hardware driver interface
• Runtime interfaces for experiments:
  • desired trajectory input
  • torque/current override input
  • online PID gain update input
  • present-current feedback publishing

Safety and Reliability Layer

• Current clamp to hard limits
• Command freshness timeouts
• Explicit operating-mode switching for motors:
  • torque off
  • set operating mode to current control
  • torque on
• Optional online current readback for monitoring

Key Outcomes

• Delivered a deployable adaptive controller on real Ned3 Pro hardware despite the lack of a direct torque command interface.
• Closed the practical gap between torque-law controller design and current-mode motor actuation.
• Built a controllable and testable ROS runtime for controller experiments with safety constraints.
• Verified this controller approach on three primary robot motors.