miniplan

A PDDL planner library built around the pddl crate.

miniplan is a grounded, classical PDDL planner focused on simplicity and performance. It targets STRIPS-level domains with support for typing, negative preconditions, and conditional effects. Use it as a library in your own projects or as a CLI (miniplan) for quick planning experiments.

Features

• PDDL 3.x parsing via pddl-rs
• Grounding to a compact bitset-based internal representation
• Planners:
  • bfs — blind breadth-first search (optimal)
  • astar — A* with pluggable heuristics (optimal)
  • gbfs — greedy best-first search (fast, non-optimal)
  • bidij — bidirectional Dijkstra (cost-aware, optimal)
  • bibfs-uc — bidirectional BFS (uniform-cost, optimal)
  • nbs — Near-Optimal Bidirectional Search (heuristic-guided, non-optimal with HFF)
  • bae — Bidirectional A* with Error / BAE* (heuristic-guided, non-optimal with HFF)
• Bidirectional heuristic search: bidij, bibfs-uc, nbs (NBS), bae (BAE*)
• Heuristics: hff (relaxed-plan), hadd, hmax, goal-count, blind
• Output formats: plain, IPC, JSON
• Search limits: time budget, node budget
• Structured logging via tracing

Installation

Library

[dependencies]
miniplan = "0.1"

CLI

cargo install miniplan-cli --locked

The resulting binary is miniplan (not miniplan-cli).

Or install from the git repository:

cargo install --git https://github.com/sunsided/miniplan miniplan-cli

Build from source

git clone https://github.com/sunsided/miniplan
cd miniplan
cargo build --release

Quick Start (CLI)

Solve a bundled example with search statistics:

$ miniplan solve --stats examples/pddl/blocksworld.pddl
; cost = 8
; length = 8
(unstack(a,c))
(put-down(a))
(pick-up(c))
(stack(c,d))
(pick-up(b))
(stack(b,c))
(pick-up(a))
(stack(a,b))
Search stats:
  Nodes expanded: 27
  Nodes generated: 45
  Plan cost: 8.00
  Plan length: 8
  Time: 3.451169ms

Parse and ground a problem without solving:

miniplan check examples/pddl/gripper.pddl

List available planners and heuristics:

$ miniplan list-planners
Available planners:
  bfs          Breadth-first search
  astar        A* search with pluggable heuristic
  gbfs         Greedy best-first search
  bibfs-uc     Bidirectional BFS (uniform-cost, not cost-aware)
  bidij        Bidirectional Dijkstra (cost-aware)
  nbs        Near-Optimal Bidirectional Search (Chen et al. 2017)
  bae        Bidirectional A* with Error (BAE*, Sadhukhan 2013)

Available heuristics:
  goal-count
  hff
  blind
  hadd
  hmax

CLI Reference

Commands

solve Flags

Exit Codes

Library Usage

use std::path::PathBuf;

use miniplan::pddl_io::load_files_named;
use miniplan::ground::ground;
use miniplan::search::{Solver, PlannerChoice, PlannerKind, PlannerConfig, SearchLimits, SearchOutcome};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Load PDDL files (one combined, or domain + problem)
    let files = [PathBuf::from("examples/pddl/blocksworld.pddl")];
    let (domain, problem) = load_files_named(&files, None, None)?;

    // Ground the task
    let task = ground(&domain, &problem)?;

    // Configure and run the solver
    let solver = Solver::new();

    let mut config = PlannerConfig::default();
    config.opts.insert("heuristic".to_owned(), "hff".to_owned());

    let choice = PlannerChoice {
        kind: PlannerKind::Astar,
        heuristic: Some("hff".to_owned()),
        config,
    };

    let limits = SearchLimits::default();
    let outcome = solver.solve_task(&task, &choice, &limits)?;

    match outcome {
        SearchOutcome::Plan(plan, stats) => {
            println!("Plan found (cost={:.2}, length={}):", stats.plan_cost, stats.plan_length);
            println!("{}", plan);
        }
        SearchOutcome::Unsolvable(_) => {
            eprintln!("Problem is provably unsolvable.");
        }
        SearchOutcome::LimitReached(_) => {
            eprintln!("Search limit reached.");
        }
    }

    Ok(())
}

See docs.rs for the full API.

Supported PDDL Features

Examples

Five PDDL problems are bundled in examples/pddl/:

Run all examples via the Taskfile:

task example        # Full demo sweep
task example:blocksworld   # Single example

Development

This project uses Task for automation.

task --list-all        # List all available tasks
task verify            # fmt + check + lint + test (full gate)
task example           # Run the full example suite

A pre-commit configuration is included (.pre-commit-config.yaml) for lint checks before commits.

Project Layout

crates/miniplan/         Core library
  src/
    ground/              PDDL grounding
    heuristic/           Heuristic implementations (h^FF, h^add, h^max, …)
    pddl_io.rs           PDDL file loading
    plan.rs              Plan representation and formatting
    search/              Planner implementations (BFS, A*, GBFS)
    task.rs              Grounded task representation
    util/                Utilities (bitsets, etc.)
bins/miniplan-cli/       CLI binary (produces `miniplan`)
examples/pddl/           Bundled PDDL problems
Taskfile.dist.yaml       Task definitions

Roadmap

• Current: grounded STRIPS, BFS/A*/GBFS, h^FF
• Planned: landmark heuristics, plan validation, more PDDL coverage (numerics, durative actions)

License

Licensed under the European Union Public Licence 1.2 (EUPL-1.2). See the SPDX entry or the official EU page for details.

Acknowledgements

• pddl-rs — PDDL parser, forked and maintained for this project.
• Jörg Hoffmann & Bernhard Nebel — The FF planning system and relaxed-plan heuristic (Hoffmann & Nebel, The FF Planning System: Fast Planning Generation, JAIR 2001).