Supaero Reinforcement Learning Initiative

View our GitHub profile

The SuReLI blog

The SuReLIb (private)

Postdoc in Reinforcement Learning for sim2real and related topics

Applications are invited for a full-time post-doctoral research position in the Supaero Reinforcement Learning Initiative, at ISAE-SUPAERO, Toulouse, France. Funding is secured for a 3 year position (the appointment can be shorter if needed) on a project which aims at reliability in sim2real transfer in RL.

General motivation for the project is sim2real transfer and real-life RL. Current research interests are listed below but the research agenda is quite open and can include new directions brought by the postdoc applicant. We are interested in designing new methods for domain adaptation/generalization in RL, to enable efficient transfer to unseen states or new MDPs. Recent work in the team on representation learning for generalization and transfer can be further developed. Robust MDPs and robust RL are also on the agenda, as well as mathematical programming formulations in RL. Finally, the project also aims at contributions to data-frugal RL. Besides these key topics, the research agenda is quite flexible and open; both theoretical and applied work is of interest. Preferred applications concern mobile robotics but extensions to other applications are possible.

Keywords: sim2real, domain adaptation/generalization, transfer, robust RL, data frugality, representation learning for generalization.

The Supaero Reinforcement Leaning Initiative (SuReLI) is a group of professors, researchers, engineers, students, researching open questions in Reinforcement Learning. We actively promote an incusive, friendly work atmosphere. Although we participe in the current race for research in ML, we are also outspoken promoters of the virtues of slow science. We are located in ISAE-SUPAERO, the world leading institution in aerospace engineering higher education, in the lively city of Toulouse, regularly ranked best student city in France. SuReLI’s recent research spans topics such as representation learning for RL [ICLR2022] [NeurIPS2022] [NNLS2023], RL in non-stationary MDPs [AAAI2021], evolutionary RL and optimization [GECCO2021] [GECCO2022], distillation of deep RL policies [CoLLAs2022], optimization for RL [ICML2022], neural architecture search [AutoML2022], and applications to robotics, fluid flow control, software testing, video games, etc.

Candidates should hold a PhD in a relevant discipline (computer science, mathematics, control theory, etc.), be scientifically curious, technically autonomous and able to suggest original research directions. Applicants should hold a good publication record at top-tier ML conferences. Group involvement and student advising within SuReLI is encouraged, so applicants should have demonstrated their advising or leadership capabilities in previous projects.

Candidates are invited to contact Emmanuel Rachelson to discuss their research project and for further details on the position and the application process.