Tower cranes are complex systems that can experience payload oscillations during rapid trolley, slewing, and hoisting movements. These oscillations can impact accuracy, increase settling time, and pose challenges for autonomous and robotic crane systems. A project is underway to develop a physics-based simulation and reinforcement-learning framework for swing-aware control of a lab-scale tower crane.
The project aims to enhance the existing crane simulation model in Isaac Sim by focusing on areas like suspended payload representation, pendulum motion, sensor integration, actuator behavior, friction, and external disturbances such as wind. The physical lab-scale crane will be utilized for model identification and simulation calibration to estimate dynamic parameters and enhance simulation fidelity.
By creating an Isaac Lab RL environment, the project aims to train a model-free controller that can effectively transport the payload towards the desired target. The robustness of the controller against variations in payload mass, cable length, friction, wind disturbances, and modeling uncertainties will be tested through domain randomization. The ultimate goal is to implement the trained policy in the physical lab-scale tower crane.
The project is divided into two main subjects:
Subject 1: Physics-Based Simulation and Model Identification
– Completing the tower crane URDF model in Isaac Sim
– Modeling crane actuation, payload dynamics, and wind disturbances
– Integrating sensors and updating the simulation to match the real crane
– Conducting experiments on the lab-scale crane to identify key dynamic parameters
– Tuning and validating the simulation model against real-world measurements
– Developing the simulation for integration with a second robotic system
Subject 2: Reinforcement Learning and Experimental Validation
– Creating the crane reinforcement-learning environment in Isaac Lab
– Defining observations, actions, reward functions, and training scenarios to optimize payload transport and minimize oscillations
– Training a model-free reinforcement learning algorithm to control the crane
Applicants are required to have proficiency in reinforcement learning, Isaac Sim/Lab, C++/Python, basic knowledge of dynamics and control theory. Experience with real robots is a plus. Interested individuals can apply by sending their CV, transcript, and any relevant experience with Isaac Sim/Lab, reinforcement learning, simulation, or robotics to mohammadreza.kolani@tum.de.
The project offers a unique opportunity to work on cutting-edge research at the intersection of robotics, simulation, and reinforcement learning. For more information and application details, visit https://www.cee.ed.tum.de/ccbe/labs/robotic-fabrication-lab/.
About the Author
