The Planner

Autoplan is the planner behind Autoplan Studio: one domain-independent engine that reads a model written in PDDL, searches for a plan, and checks that plan before reporting it.

Autoplan was developed by Josef Bajada who holds a Ph.D. in Computer Science specialising in AI Planning from King's College London, together with years of experience building mission critical systems, including proprietary planners, for industry.

Autoplan handles classical, numeric and temporal planning in one program. A model can use plain facts, quantities such as fuel or stock, and actions that take time, overlap and change those quantities while they run. The supported features page lists the PDDL features it covers. If you are new to the field, What is AI Planning? explains the ideas.

From research to industry

Numeric and temporal planning are hard for different reasons. Quantities give a search far more states to tell apart, and actions that must run together have to be scheduled as well as chosen. Autoplan builds on the research literature on both, including the author's own published work on temporal planning with continuous change.

Developing planners for industry shaped its priorities as much as the research did. Real models come with deadlines, and the people using them need to trust the answer. So Autoplan is built to:

  • reach a first plan quickly, and keep memory use visible on every run;
  • validate every plan against the full model before calling it a plan;
  • report anything it cannot plan with, rather than silently ignoring it;
  • run as one self-contained binary, written in Rust from the parser to the validator.

Some of the techniques in Autoplan are still unpublished but its results can still be examined. The benchmarks compare it with 14 other planners on International Planning Competition problems, counting a plan only when the independent VAL validator accepts it.

Planner availability

The planner is available to try for free via the Autoplan Studio web-based user interface. If you are interested in using the planner commercially, feel free to get in touch.

Publications

  1. Josef Bajada. Tempest Numeric: A Planner for the IPC 2026 Numeric Track. Planner abstract, IPC 2026 numeric track, 2026.
  2. Josef Bajada, Maria Fox, Derek Long. Efficient Temporal Piecewise-Linear Numeric Planning With Lazy Consistency Checking. IEEE Transactions on Artificial Intelligence, 3(4), 506-517, 2022. Preprint.
  3. Josef Bajada, Maria Fox, Derek Long. Temporal Planning with Constants in Context. ECAI 2016, Frontiers in Artificial Intelligence and Applications 285, 1712-1713, IOS Press, 2016.
  4. Josef Bajada, Maria Fox, Derek Long. Temporal Planning with Semantic Attachment of Non-Linear Monotonic Continuous Behaviours. IJCAI 2015, 1523-1529, 2015.