DC5 - Reliable Integration of Safety-Constraints into the Learning Process of Collaborative Robots

Objectives

1. To develop a conceptual framework that integrates modern RL and IL approaches for acquiring complex robotic manipulation skills in collaborative settings while ensuring compliance with functional safety regulations. The framework will incorporate validity concepts for machine learning methods and utilize safety techniques such as shielding to meet safety-critical requirements.
2. To integrate a fast, reactive digital twin (DT), leveraging co-simulation to anticipate and prevent unsafe scenarios in real-time, enabling proactive behavioural adjustments and sim-to-real skill development.

Expected Results

1. A framework based on RL and IL that is conform with functional safety standards will provide domain experts with easy-to-use tools to develop specific robotics applications in the field of manufacturing that are safe during the whole life-cycle of the focused collaborative robotics task.
2. Validation of the performance and safety of the digital twin of the developed framework for the specific application. Since all skills can be acquired in simulation, performance and safety should be measurably improved over frameworks ignoring the benefits of physical simulations.

Planned Secondment(s)

S9: LNE, Dr. Kalouguine, 2 months in M14-15
S10: PAL, Dr. Lemaignan, 2 months in M26-27

Required Skills

Essential

• Computer science/engineering degree
• Excellent programming skills (e.g. Python)

Desirable

• Practical experiences with (collaborative) robotics and safety
• Knowledge in ROS and robot simulators (e.g. Gazebo, IsaacSim, etc.)
• Practical experience with machine learning algorithms and developing/testing of complex software systems
• Basic knowledge of human-robot interaction