Halim Djerroud

My research is grounded in the idea that an autonomous system should not only be efficient, but also understand what it does, why it does it, and when it should change its decision-making strategy. I approach autonomy through the lens of embodied decision-making, where a robot's choices are constrained by its body, its sensors, its action capabilities, and the physical and social environment in which it operates.

Inspired by cognitive sciences, I consider decision-making as a fundamentally prospective process: acting requires anticipating the consequences of actions, relying on internal simulation, memory of past successes and failures, and structured abstractions of the environment. Rather than pursuing autonomy through ever-increasing model complexity or computational power, my work explores architectures that explicitly reason, simulate possible futures, and adapt their decision processes over time.

A central aspect of this approach is cognitive frugality. An intelligent system should not only decide what to do, but also how much reasoning is necessary to do it, and when it is preferable to reuse experience, simplify its reasoning, or involve a human. By integrating uncertainty as a core component of decision-making rather than a limitation to eliminate, my research aims to design autonomous robotic systems that are more explainable, adaptive, and robust, and better suited to dynamic, shared, and real-world environments.