Optimization of sim-to-real transfer in the humanoid robot NICO

Anotation

The growing demand for intelligent and adaptive robots has led to significant advances in simulation, control, and learning algorithms. However, transferring behaviors learned in simulation to the real world remains a major challenge due to discrepancies between simulated and physical dynamics, a phenomenon known as the reality gap. This thesis focuses on improving the sim-to-real transfer in the humanoid robot NICO, with an emphasis on reaching and grasping tasks.

Thesis objectives

1. Using the MyGym robotic simulator, which includes a model of the humanoid robot NICO, design a calibration method that allows the simulator to be linked to a physical robot.
2. The method will consist of measuring the positions of paired points (in 3D), from which a mathematical representation (e.g., using a neural network) between the two spaces will be created.
3. Test the accuracy of the proposed method using appropriate motor operations of the robot arm.