PhD Position F/M Enforcing Persistence in Generative World Models for Robotics

Context
AI4IDF is a structuring and unifying research project dedicated to artificial intelligence, a scientific and technological field that has become essential.
Aiming to deepen knowledge in AI while keeping human beings at the center of its concerns, the Ile-de-France Region aspires to play a major role in this promising sector, through its scientific excellence and its incomparable ecosystem, and wishes to make the latter a key player in building a frugal, reliable, and efficient Artificial Intelligence (AI), designed to coexist with humans and assist them in their decisions.
To this end, DIM AI4IDF periodically launches calls for doctoral research projects. This thesis falls within this framework and establishes a partnership between the Hi!Paris and PRAIRIE clusters, specifically between Telecom ParisTech and Inria Paris Centre.
The objective of this thesis is to conduct joint research on persistence in generative world models for robotics and to develop algorithms capable of integrating physical concepts during image generation in particular. The project will develop methods to enforce persistence in generative world models, with contributions including a formal definition of persistence, trajectory-level learning methods, and empirical validation in robotic prediction and control.
Regular travel is expected for this position, notably for dissemination purposes, scientific training, and participation in project and collaboration meetings.
Assignment
Responsibilities:
Under the guidance of researchers from Inria and Telecom ParisTech, the successful PhD candidate will carry out scientific research and software development under the direct supervision of their supervisors.
For further information on the proposed research topic:
A state-of-the-art review, a bibliography and scientific references are available at the following URL; please feel free to visit the site:
Collaboration:
The successful candidate will be linked to the research and training ecosystems of Telecom ParisTech and the Paris centre of Inria. In particular, the Multimedia (MM) laboratory at Telecom ParisTech (*) and the ASTRA project team at Inria Paris (**).

• https://www.telecom-paris.fr/en/research/labs/information-processing-ltci/teams/multimedia

** https://astra-team.github.io
Main activities
Main activities:
Conducting research and development
Writing scientific articles and reports
Contribute to the dissemination and popularisation of scientific knowledge
Undertake scientific and professional development training
Participate in the activities of their research teams and in events organised by their laboratories
Other activities:
Participate in contractual activities and collaborative projects undertaken by their respective teams.
Skills
Technical skills and level required:
Languages: proficiency in English (written and spoken) is required. French is strongly recommended.
Interpersonal skills: being open to multicultural and inclusive research environments and adhering to the social and professional code of conduct established in the host laboratories. Respecting professional and conduct rules as well as research ethics is required.
Additional skills appreciated: proficiency in tools, languages, and development libraries for computer vision and AI development. The essential combination is Python, PyTorch, Git (Linux is a plus), complemented by skills in GPU computing, large-scale data processing, and a solid mathematical background.
Specifically for the subject of this thesis, knowledge in the following areas is strongly recommended:
Generative models:
diffusion
(Stable
Diffusion,
DDPM),
flows,
VAE.
Frugal AI:
quantization,
pruning,
distillation,
edge
computing
(TensorRT,
ONNX).