2R Robot Arm Lab Experiment – Gravity Compensation and Reinforcement Learning
Abstract
The M-IAS (Intelligent Autonomous Systems) master’s programme at TH Nürnberg requires a new lab experiment for the “Intelligent Robotics” practical course that demonstrates fundamental concepts of control theory and machine learning in a hands-on setting. The goal of this thesis is to build a two-link robot arm (2R kinematics) based on two BLDC motors, serving as a versatile demonstration platform for gravity compensation and reinforcement learning.
On the hardware side, both joints are to be equipped with BLDC motors and suitable motor drivers (e.g. ODrive or VESC). The mechanical structure of the arm segments is to be designed and manufactured, considering stiffness, weight and accessibility for students. Encoders provide the necessary position and velocity feedback for control.
On the software side, forward and inverse kinematics are to be implemented first, along with model-based gravity compensation that compensates for the weight of the arm segments in real time, keeping the arm floating in any position. This physically intuitive experiment forms the entry point for students into model-based control.
Building on this, a reinforcement learning demo is to be developed where an agent learns to move the arm to a target position. A ROS-based system architecture is to be established supporting both real hardware and a simulation environment (e.g. Gazebo or MuJoCo) for RL training. Lab instructions for student use in M-IAS are to be prepared.
Motivation
The “Intelligent Robotics” practical course in the M-IAS master’s programme currently lacks a hands-on experiment bridging classical model-based control and modern learning-based methods. A 2R robot arm with torque control offers exactly this: students first experience the physical effects of gravity on the real system, then learn how an RL agent solves the same task in a data-driven way.
Work Packages
Hardware
- Selection and commissioning of BLDC motors and motor drivers (e.g. ODrive / VESC)
- Mechanical design and fabrication of arm segments (CAD, 3D printing or aluminium profile)
- Integration of encoders and wiring
Software & Control
- Implementation of forward and inverse kinematics
- Model-based gravity compensation controller (real-time torque compensation)
- ROS integration: topics, services and ros2_control
Reinforcement Learning
- Building a simulation environment (Gazebo or MuJoCo) for RL training
- Training an agent (e.g. PPO / SAC) on a target-reaching task
- Sim-to-real transfer and evaluation on the real arm
Documentation
- Creating lab instructions for use in the M-IAS practical course
Requirements
- Knowledge of control theory and robotics (kinematics, dynamics)
- Programming skills in Python and/or C++
- Ideally experience with ROS
- Interest in machine learning and reinforcement learning
- Enthusiasm for hands-on hardware work and prototyping
This topic can be completed as a project or master’s thesis subject to agreement.
Supervision
| Role | Name | |
|---|---|---|
| Supervisor | Prof. Dr. Christian Pfitzner | christian.pfitzner@th-nuernberg.de |
Prof. Dr. Christian Pfitzner
Professor of Mobile Robotics
Professor of Mobile Robotics at TH Nürnberg Georg Simon Ohm, Faculty efi. Director of the TTZ Nürnberger Land. Research interests in autonomous navigation, drone technology and sensor fusion.