Image: Philip Liljenberg, Johnér Bildbyrå AB
Image: Philip Liljenberg, Johnér Bildbyrå AB
Research project In this research project, a physics-informed digital twin of a rock breaker is developed and made autonomous using deep learning.
The goal is to create physics-informed autonomous manipulation of granular matter. The idea is to learn a dynamics model and the current state of fragmented material in the environment of a mechanical manipulator. The learning takes place in a virtual environment where a wide range of different materials and states can be represented, as well as different manipulators and various sensors for creating synthetic training data.
The goal of the project is to develop and explore: i) a digital twin of a rock breaker, that is, a simulator that accurately captures the physics of breakers and rocks, and that allows for rendering synthetic sensor data and machine learning through massive simulations on a high-performance computing cluster. ii) different levels of autonomous control of rock breakers through physics-informed deep neural networks trained in a virtual environment. iii) solutions for transferring observations in the physical domain to the virtual one (real2sim) and of trained neural networks predicting actions from the simulated to physical domain (sim2real).
