Frontmatter

Roman Barták, Andrea Orlandini, Mauro Vallati, Sven Koenig, Roni SternProceedings of the International Conference on Automated Planning and Scheduling 2023 · DOI

Abstract: This volume contains the papers accepted for presentation at ICAPS 2023, the Thirty-Third International Conference on Automated Planning and Scheduling, held in Prague, Czech Republic, July 8-13, 2023. The annual ICAPS conference series was formed in 2003 through the merger of two pre-existing biennial conferences, the International Conference on Artificial Intelligence Planning and Scheduling (AIPS) and the European Conference on Planning (ECP). ICAPS continues the traditional high standards of AIPS and ECP as an archival forum for new research in the field of automated planning and scheduling.

Introduction

The field of Automated Planning and Scheduling (APS) is at a critical juncture, balancing the rigorous foundations of symbolic logic with the rapid advancements of modern machine learning. The ICAPS 2023 proceedings represent a comprehensive overview of this evolution, highlighting how autonomous systems are moving from controlled laboratory environments into the "wild"—complex, real-world settings such as smart cities, digital forensics, and collaborative robotics.

The primary challenge addressed by the 2023 cohort is the need to scale autonomous decision-making systems while maintaining the reliability and "explainability" inherent in traditional planning methods. As these systems take on higher stakes, the community is increasingly focused on bridging the gap between theoretical analysis and empirical performance in unpredictable environments.

While the ICAPS proceedings cover a diverse array of research, several dominant methodological trends emerged during the 2023 conference:

  • Heuristic Search and Pruning: Research continues to refine the efficiency of searching massive state spaces. Key advancements include "Landmark Progression" and "Lifted Mutex Groups," which allow planners to navigate complex problems by identifying critical milestones and structural constraints without exhaustive state exploration.
  • Hybrid Learning-Planning: A significant portion of the program was dedicated to the intersection of Planning and Learning (PRL). This track explores how Reinforcement Learning (RL) and Deep Learning can be used to either enhance traditional planning heuristics or replace them entirely in specific contexts.
  • Formal Logic and Temporal Constraints: The use of Pure-Past Linear Temporal Logic (PPLTL) has gained traction for managing temporally extended goals, allowing systems to reason about history and time-sensitive requirements more effectively.
  • Unified Frameworks: To lower the barrier to entry for non-experts, the community introduced the "Unified Planning" Python library. This initiative aims to provide a standardized interface for various planning engines, making sophisticated APS technology accessible to engineers in adjacent fields.

Key Insights and Emerging Themes

The Pivot to Explainability

A central theme of ICAPS 2023 was the shift toward human-aware AI. Through invited talks and dedicated workshops (such as HAXP), the community emphasized that it is no longer enough for a system to find an optimal plan; it must also be able to explain why that plan was chosen. This move away from "black box" autonomy is essential for building trust in human-robot collaboration.

The Role of Large Language Models (LLMs)

The 2023 conference marked a pivotal moment with the inclusion of tutorials on LLMs in planning. Researchers are beginning to evaluate how generative AI can assist in model acquisition—the process of defining the rules of a world—and how LLMs might assist in high-level plan generation, despite their current limitations in logical reasoning.

Socially Impactful Applications

The broadening scope of APS was exemplified by the "Innovative Paper Award," which recognized research applying automated planning to digital forensics. By using planning algorithms to prioritize the investigation of indecent images of children, the research demonstrated how APS can be applied to high-stakes, socially critical decision-making processes.

Conference Outcomes and Results

ICAPS 2023 maintained a high standard of selectivity, with an acceptance rate of approximately 30% (79 papers accepted from 258 submissions). The following papers were recognized for their exceptional contributions:

  • Best Paper: "Landmark Progression in Heuristic Search" by Büchner et al., which advanced the state-of-the-art in search efficiency.
  • Best Student Paper: "Planning for Temporally Extended Goals in Pure-Past Linear Temporal Logic" by Bonassi et al., highlighting the importance of temporal reasoning in modern planners.

Beyond the main track, the conference featured 12 workshops and 6 tutorials, covering specialized domains ranging from financial services (FinPlan) to robotics (PlanRob).

Limitations and Future Directions

Despite the progress showcased in Prague, the proceedings identify several ongoing hurdles for the APS community:

  1. Integration Gaps: There remains a significant need for better integration between AI Planning and Reinforcement Learning. These sub-fields often operate with different assumptions and terminologies, hindering the development of truly hybrid systems.
  2. Model Acquisition Bottlenecks: Creating the initial "world model" for a planner remains a manual and labor-intensive process. Future work must focus on automated model acquisition to allow planners to adapt to new domains more quickly.
  3. Accessibility: While the "Unified Planning" library is a step in the right direction, the barrier to entry for implementing these algorithms in industrial settings remains high. Simplifying the deployment of these complex systems is a priority for the coming years.

Summary updated

Abstract

This volume contains the papers accepted for presentation at ICAPS 2023, the Thirty-Third International Conference on Automated Planning and Scheduling, to be held in Prague, Czech Republic, July 8-13, 2023. The annual ICAPS conference series was formed in 2003 through the merger of two pre-existing biennial conferences, the International Conference on Artificial Intelligence Planning and Scheduling (AIPS) and the European Conference on Planning (ECP). ICAPS continues the traditional high standards of AIPS and ECP as an archival forum for new research in the field of automated planning and scheduling.
BibTeX
@inproceedings{bartak2023frontmatter,
  author = {Roman Barták and Andrea Orlandini and Mauro Vallati and Sven Koenig and Roni Stern},
  title = {{Frontmatter}},
  booktitle = {Proceedings of the International Conference on Automated Planning and Scheduling},
  year = {2023},
  doi = {10.1609/icaps.v33i1.27251},
}
Figure