Internship: Learning-Based Optimisation Of Multi-Robot Systems F/M

With the growing development of robotics services, the problem of orchestrating a fleet of robots (or autonomous agents) under various constraints has recently become a major design bottleneck, especially when seeking to optimise service operations. In the Optimization with Learning team, we are interested in optimising pick-up and delivery services involving robot fleets moving in open (indoor or outdoor) environments. The underlying challenges stem from hard combinatorial optimisation problems, such as multi-robot routing and scheduling under uncertainty.

This internship is related to our research on Neural Combinatorial Optimization for Robot Fleet Management. More information about this research can be found here: https://europe.naverlabs.com/research/neural-combinatorial-optimization-robot-fleet-management/

At least two broad approaches traditionally address this type of problem, each with its advantages and drawbacks, especially in the face of uncertainty. On the one hand, Reinforcement Learning (and more generally Sequential Decision Processes) attempts to predict the optimal action at any given instant, based on past or simulated experiences. On the other hand, multi-agent planning aims to build optimal plans given a model of the environment.

The purpose of the internship is to:

• explore the space at the intersection of these two traditions and review the existing literature,

• design new learning algorithms that capture the best of both worlds, and

• conduct experiments to evaluate such algorithms on some of our multi-robot service use cases in simulated environments.

About the research team

In the Action group, we develop AI-driven decision-making capabilities that enable embodied agents to safely execute complex tasks in dynamic environments.To achieve autonomy in real-world everyday spaces, robots must learn from their interactions, understand how to best execute tasks specified by non-expert users, and do so in a safe and reliable manner. This requires sequential decision-making skills that integrate machine learning, adaptive planning, and control in uncertain environments, as well as the ability to solve hard combinatorial optimization problems. Our research combines expertise in reinforcement learning, computer vision, robotic control, sim-to-real transfer, large multimodal foundation models, and neural combinatorial optimization to design AI-based architectures and algorithms that enhance robot autonomy and robustness when completing complex tasks in constantly changing environments.

What we're looking for

• Enrolment in a PhD or Master's program in Machine Learning or Computer Science

• Familiarity with Machine Learning for graph data

• Hands-on experience with Python and PyTorch

• Interest in combinatorial optimisation and its applications

• Understanding of Reinforcement Learning and planning is a plus

Practical Information

Start date: As soon as possible Duration: 6 months