Learning world models for advanced autonomous agent
World models are internal representations of the external environment that an agent can use to interact with the real world. They are essential for understanding the physics that govern real-world dynamics, making predictions, and planning long-horizon actions. World models can be used to simulate real-world interactions and enhance the interpretability and explainability of an agent's behavior within this environment, making them key components for advanced autonomous agent models.
Nevertheless, building an accurate world model remains challenging. The goal of this PhD is to develop methodology to learn world models and study their use in the context of autonomous driving, particularly for motion forecasting and developing autonomous agents for navigation.
Accelerating thermo-mechanical simulations using Neural Networks --- Applications to additive manufacturing and metal forming
In multiple industries, such as metal forming and additive manufacturing, the discrepancy between the desired shape and the shape really obtained is significant, which hinders the development of these manufacturing techniques. This is largely due to the complexity of the thermal and mechanical processes involved, resulting in a high computational simulation time.
The aim of this PhD is to significantly reduce this gap by accelerating thermo-mechanical finite element simulations, particularly through the design of a tailored neural network architecture, leveraging theoretical physical knowledge.
To achieve this, the thesis will benefit from a favorable ecosystem at both the LMS of École Polytechnique and CEA List: internally developed PlastiNN architecture (patent pending), existing mechanical databases, FactoryIA supercomputer, DGX systems, and 3D printing machines. The first step will be to extent the databases already generated from finite element simulations to the thermo-mechanical framework, then adapt the internally developed PlastiNN architecture to these simulations, and finally implement them.
The ultimate goal of the PhD is to demonstrate the acceleration of finite element simulations on real cases: firstly, through the implementation of feedback during metal printing via temperature field measurement to reduce the gap between the desired and manufactured geometry, and secondly, through the development of a forging control tool that achieves the desired geometry from an initial geometry. Both applications will rely on an optimization procedure made feasible by the acceleration of thermo-mechanical simulations.