Pylic: Leveraging Source Code for Planning in Structured Environments

Leonardo Hernandez Cano, Nathan Hunt, Sara Magliacane, Armando Solar-Lezama

Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: program analysis, planning, robotics, optimization

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We leverage simulator source code structure to describe and solve the combined planning and control problem

Abstract: This paper investigates the application of program analysis techniques to planning problems in dynamic environments with discontinuities in long-horizon settings. Traditional approaches rely on specialized representations, which are often tailored to specific problems and domains. In contrast, we propose describing the combined planning and control problem directly as a desired property of the execution of simulator source code. This representation is expressive, naturally providing a means to describe desired properties of even very dynamic and discontinuous environments. We show that, despite this generality, it is still possible to leverage domain knowledge by relating it to the simulator source code. We study the effectiveness of this approach through several case studies in simulated robotic environments. Our results show that in these environments, our framework can improve the efficiency in solving the control and planning problem, relative to standard numerical and reinforcement learning methods.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6159