Marvin* is a research group of the Laboratoire d'Informatique de Grenoble that designs AI planning algorithms for decision-making in autonomous systems, with diverse applications in robotics and Industry 4.0. In artificial intelligence, automated planning aims to develop algorithms capable of generating action plans, typically intended to be executed by a robot or other autonomous agents. The knowledge and actions that agents can perform are modeled using domain-independent languages, such as PDDL (Planning Domain Definition Language). The complexity of the planning problem varies depending on the simplifying assumptions adopted, such as accounting for time, the nondeterministic nature of actions, or the partially or fully observable environment.
The main challenges that the Marvin team strives to address include:
- Developing and learning heuristic functions to improve the efficiency of finding solution plans.
- Creating explanatory or collaborative planners capable of justifying their decisions or generating plans interactively with humans.
- Designing algorithms that generate expressive plans, incorporating properties such as concurrency, loops, or complex conditions.
- Developing closed-loop planners that integrate planning, execution, and replanning processes.
- Learning and modeling agents' knowledge from execution traces of existing systems.
Through these efforts, the Marvin team actively contributes to pushing the boundaries of automated planning while promoting its integration into complex real-world contexts.
(*) Marvin refers to Marvin Minsky, one of the founding fathers of Artificial Intelligence, but also to the fictional robot Marvin, the ship's robot in The Hitchhiker's Guide to the Galaxy series by Douglas Adams.
###