Featured Software & Tools
Open-source planners and research tools
K* Search Planners
K* search based planners for top-k and top-quality planning tasks. Pip-installable for easy integration.
Forbid-Iterative (FI)
Planner suite for top-k, top-quality, and diverse computational tasks in automated planning.
Diversity Score
Tool for computing diversity scores for sets of plans, enabling quality assessment of diverse planning.
Cerberus Planner
Post-IPC 2018 version of the Cerberus planner, successor to Mercury with advanced red-black heuristics.
AI Planning Service
IBM Research AI Planning Service for integrating automated planning into applications and workflows.
Competition Winners
Award-winning planners from international competitions
π₯ IPC 2018 - Delfi Planner
Winner of the Sequential Optimal Track
Delfi won the cost-optimal track of the International Planning Competition 2018, demonstrating state-of-the-art performance in finding optimal solutions.
π₯ IPC 2014 - Mercury Planner
Runner-Up & Innovative Planner Award
Mercury achieved runner-up in the satisficing track and received the Innovative Planner Award at IPC 2014 for its novel red-black planning approach.
Other IPC 2018 Planners
IPC 2014 Planners
Patents
25+ granted and published patents
Granted Patents
-
Estimating and visualizing collaboration to facilitate automated plan generation
US Patent 11620486 -
Methods and systems for diverse instance generation in artificial intelligence planning
US Patent 11526791 -
Guided plan recognition
US Patent 11755923 -
Iterative generation of top quality plans in automated plan generation
US Patent 11727289 -
Optimizing spatiotemporal computational problems
US Patent 10635982 -
Automatic solution to a scheduling problem
US Patent 10430739 -
Reusable modeling for solving problems
US Patent 10169291
Recent Published Patents
-
Partial Order Reduction to Increase Planner Speed
US 20240420020 -
Symmetry Pruning to Increase Planner Speed
US 20240420038 -
Reinforcement Learning Using Lifted Action Models
US 20240370750 -
Generating Artificial Intelligence Plans of High Diversity
US 20230394325 -
Action Space Reduction for Planning Domains
US 20230342653
Research Highlights
Key contributions and innovations
π€ Thought of Search (ToS)
Novel approach for planning with language models through the lens of efficiency.
π§ Generalized Planning with LLMs
Improved generalized planning via pseudocode strategy refinement and reflection, achieving 82% average coverage across 17 PDDL domains.
π Fact-Level Relevance Pruning
Aggressive task simplification by reasoning about relevance at the level of facts rather than variables, preserving all shortest optimal plans.
π‘ GNN-Based Planner Selection
Heterogeneous Graph Neural Networks (RGCN+XGBoost) for online planner selection, achieving 91.7% accuracy β improving over prior methods.
π’ LLM-Based Planning: Position Papers
Two position papers on the state and future of planning in the LLM era: examining planner-generation methods (NL2Search, NL2PDDL, NL2Policy) and calling for rigorous methodology grounded in six decades of planning community insights.
π ACPBench
Comprehensive benchmark for reasoning about action, change, and planning with language models.
π Diverse Planning
Techniques for generating diverse sets of plans, providing users with meaningful alternatives and flexibility in decision-making.
π― Top-Quality Planning
Methods for finding practically useful sets of best plans, enabling better decision-making in planning applications.
π΄β« Red-Black Planning
Systematic approach to partial delete relaxation, forming the basis of award-winning planners Mercury and Cerberus.
π Planning & RL
Bridging automated planning and reinforcement learning for autonomous systems and hybrid approaches.
Collaborators
Working with leading researchers worldwide
I have had the privilege of collaborating with outstanding researchers from academia and industry, including colleagues from IBM Research, MIT, Technion, University of Basel, Saarland University, and many other institutions.