Software

Libraries

Mimir: A Generalized Planning Library

• Mimir is a C++20-based generalized planning library with Python bindings, designed for both grounded and lifted planning. Its primary focus lies at the intersection of planning and machine learning, making it a powerful tool for research in learning-based planning. Unlike many planners, Mimir is designed to be used as a library rather than as a standalone application, offering seamless integration into both C++ and Python projects. Pre-compiled Python binaries are available via PyPI for easy installation and use.
• Features
  • Rich PDDL Support
  • Grounded and Lifted Planning
  • Collections of Problems
  • Portabile, Flexible, Efficient

Description Logics State Features for Planning Library (DLPlan)

• The library consists of five components. Each component has its own public header file, examples, tests, and python bindings.
  • Core - The core component provides functionality for the construction an evaluation of domain-general state features based on description logics.
  • Generator - The generator component provides functionality for automatically generating a set of domain-general state features that are distinguishable on a given finite set of states.
  • Policy - The policy component implements the general policy language.
  • State Space - The state space provides functionality for generating state spaces from PDDL.
  • Novelty - The novelty component provides functionality for width-based planning and learning.

Softwares

Learning General Policies From Examples

• Authors: Blai Bonet, Hector Geffner
• Conference: Upcoming
• Date: 2025
• Software

Learning Lifted STRIPS Models from Action Traces Alone: A Simple, General, and Scalable Solution

• Authors: Jonas Gösgens, Niklas Jansen, Hector Geffner
• Conference: International Conference on Automated Planning and Scheduling (ICAPS) 2025
• Date: 2025
• Software, Software: Implementation of the SIFT algorithm.

Learning More Expressive General Policies for Classical Planning Domains

• Authors: Simon Ståhlberg, Blai Bonet, Hector Geffner
• Conference: Advancement of Artificial Intelligence (AAAI) 2025
• Date: 2025
• Software: Implementation of R-GNN[t] and Edge Transformers (ET) on domains such as Blocks, Grid, Gripper, Logistics, Miconic, Rovers, Vacuum, and Visitall.

Learning to Ground Existentially Quantified Goals

• Authors: Martin Funkquist, Simon Ståhlberg, Hector Geffner
• Preprint: arXiv
• Date: 2024
• Software: Contains code for Relational Graph Neural Networks (R-GNNs). Domains include Blocks, Gripper, Delivery, and Visitall from IPC but augmented with colors.

Symmetries and Expressive Requirements for Learning General Policies

• Authors: Dominik Drexler, Simon Ståhlberg, Blai Bonet, Hector Geffner
• Conference: International Conference on Principles of Knowledge Representation and Reasoning (KR) 2024
• Date: 2024
• Software: Symmetry detection in generalized planning. Benchmark set consists of domain and instances from International Planning Competition (IPC). Uses Mimir and nauty.

Expressing and Exploiting Subgoal Structure in Classical Planning Using Sketches

• Authors: Dominik Drexler, Jendrik Seipp, Hector Geffner
• Conference: Journal of Artificial Intelligence Research (JAIR) 2024
• Date: 2024
• Software: Contains code to build LAMA, Dual-BFWS, SIW and SIW_R planners. Benchmark set consists of tasks from satisficing track of IPC and Autoscale benchmark set. Uses Mimir and DLPlan.

On Policy Reuse: An Expressive Language for Representing and Executing General Policies that Call Other Policies

• Authors: Blai Bonet, Dominik Drexler, Hector Geffner
• Conference: International Conference on Automated Planning and Scheduling (ICAPS) 2025
• Date: 2024
• Software: The implementation of the SIW_M interpreter and all extended, indexical policies and sketches, and modules discussed in the paper.

Learning generalized policies for fully observable non-deterministic planning domains

• Authors: Till Hofmann, Hector Geffner
• Conference: International Joint Conference on Artificial Intelligence (IJCAI) 2024
• Date: 2024
• Software, Software: Learning general policies over fully observable, non-determinisitc (FOND) domains. Uses pddl and DLPlan.

General and Reusable Indexical Policies and Sketches

• Authors: Blai Bonet, Dominik Drexler, Hector Geffner
• Workshop: NeurIPS 2023 Workshop on Generalization in Planning (GenPlan 2023)
• Date: 2023
• Software: Implementation of the SIW_R and the SIW_M algorithms, the extended sketches, and modules presented in the article. Also contains a few example benchmarks to execute the algorithms with the extended sketches and modules. Uses Mimir.

Learning General Policies with Policy Gradient Methods

• Authors: Simon Ståhlberg, Blai Bonet, Hector Geffner
• Conference: International Conference on Principles of Knowledge Representation and Reasoning (KR) 2023
• Date: 2023
• Software: Code for training and testing the generalization, coverage, and quality of the plans obtained by the learned policies on domains such as Blocks, Delivery, Grid, Gripper, Logistics, Miconic, Reward, Spanner and Visitall.

Learning Hierarchical Policies by Iteratively Reducing the Width of Sketch Rules

• Authors: Dominik Drexler, Jendrik Seipp, Hector Geffner
• Conference: International Conference on Principles of Knowledge Representation and Reasoning (KR) 2023
• Date: 2023
• Software, Software: Hierarchical policy learner. Uses DLPlan for learning.

Learning Generalized Policies without Supervision Using GNNs

• Authors: Simon Ståhlberg, Blai Bonet, Hector Geffner
• Conference: International Conference on Principles of Knowledge Representation and Reasoning (KR) 2022
• Date: 2022
• Software: Learning generalized using graph neural networks (GNNs) from small instances in lifted STRIPS. Results on domains such as Blocks, Delivery, Gripper, Logistics, Miconic, Reward, Spanner, and Visitall.

Learning Sketches for Decomposing Planning Problems into Subproblems of Bounded Width: Extended Version

• Authors: Dominik Drexler, Jendrik Seipp, Hector Geffner
• Conference: International Conference on Automated Planning and Scheduling (ICAPS) 2022
• Date: 2022
• Software, Software: Learning policy sketches for classical planning domains and comparing the planners (1) first iteration of LAMA, (2) Dual-BFWS, (3) Serialized Iterated Width (SIW), and (4) Serialized Iterated Width with Sketches (SIWR) on a subset of classical planning benchmarks.

Learning General Optimal Policies with Graph Neural Networks: Expressive Power, Transparency, and Limits

• Authors: Simon Ståhlberg, Blai Bonet, Hector Geffner
• Conference: International Conference on Principles of Knowledge Representation and Reasoning (KR) 2022
• Date: 2022
• Software: Contains code, data and trained models of graph Neural Networks (GNNs) for learning optimal general policies over many tractable planning domains.

Flexible FOND Planning with Explicit Fairness Assumptions

• Authors: Ivan D. Rodriguez, Blai Bonet, Sebastian Sardina, Hector Geffner
• Conference: International Conference on Automated Planning and Scheduling (ICAPS) 2021
• Journal: Journal of Artificial Intelligence Research (JAIR) 2022 [Extended version of ICAPS-21 paper]
• Date: 2021
• Software: FOND-ASP planner. Solves FOND+ planning problems: FOND problems with explicit conditional fairness conditions. Includes dual FOND problems (fair and unfair actions) and QNP. Written in Answer Set Programming (ASP) using ASP Clingo system.

Learning General Planning Policies from Small Examples Without Supervision

• Authors: Guillem Francès, Blai Bonet, Hector Geffner
• Conference: Advancement of Artificial Intelligence (AAAI) 2021
• Date: 2021
• Software, Software: Implementation of D2L and evaluation on problems with simple goals such as clearing a block or stacking two blocks in Blocksworld, and standard PDDL domains such as Gripper, Spanner, Miconic, Visitall and Blocksworld.

Learning First-Order Representations for Planning from Black Box States: New Results

• Authors: Ivan D. Rodriguez, Blai Bonet, Javier Romero, Hector Geffner
• Conference: International Conference on Principles of Knowledge Representation and Reasoning (KR) 2021
• Date: 2021
• Software: Learning first-order STRIPS models from graph-based representation of state spaces

Learning First-Order Symbolic Representations for Planning from the Structure of the State Space

• Authors: Blai Bonet, Hector Geffner
• Conference: European Conf. on Artificial Intelligence (ECAI) 2020
• Date: 2020
• Software: Learning first-order STRIPS models from graph-based representation of state spaces

Qualitative Numeric Planning: Reductions and Complexity

• Authors: Blai Bonet, Hector Geffner
• Journal: Journal of Artificial Intelligence Research (JAIR) 2020
• Date: 2020
• Software: Reduction from QNPs to FOND problems.