Developing cooperative agents with a Probabilistic Programming Language


thumbnail

Suitable for:
Student, Bachelor or Master project

Description:

Probabilistic programming languages (PPLs) combine formal descriptions of computations with uncertain knowledge and are of strong interest for artificial intelligence and cognitive science researchers. Many different PPLs have been developed in recent years (e.g. http://webppl.org/, https://pyro.ai/ and others). PPLs are very powerful when dealing with uncertainties in a formal and structured way, allowing to express complex probabilistic relationships in a declarative manner. One scenario with a lot of inherent uncertainty pertains to cooperation with other agents, as their plans and intentions can only be inferred probabilistically from their behavior. The idea of this thesis topic is for a student to develop an intelligent cooperative agent in the Overcooked domain, i.e. that can cooperate with another another agent to prepare dishes in a kitchen environment. The simulated environment will be provided. The project will include implementing the necessary probabilistic inferences for behavior/intention recognition and action selection in a suitable PPL.


Contact:
Jan Pöppel