Planet: A Unified Sampling-Based Approach to Integrated Task and Motion Planning

This is a proof-of-concept implementation of the ISRR 2019 paper A Unified Sampling-Based Approach to Integrated Task and Motion Planning.

Fetching and Building

This code is designed to be easy (relative to other research code) to install and use. To get it running, you have two options:

Docker

You can use the Dockerfile in this repo to get a Docker image with Planet and its dependencies installed.

Native

To build Planet natively, follow these steps:

1. Install the following dependencies (and their dependencies):

• meson
• luajit
• A reasonable C++17 compiler (this was built using Clang++ 10.0.0)
• spdlog
• cxxopts
• boost
• OpenBLAS
• FunctionalPlus
• cpptoml
• fmt
• sexpresso
• tinyobjloader
• ompl
• bullet
• nlopt
• urdf
• tinyxml2
• Eigen
• nlohmann_json

2. Run git clone https://github.com/wbthomason-robotics/planet && cd planet && git submodule update --init --recursive to get this code and its dependent submodules.
3. Run ./build.sh {debug, debugopt, release} to build the various configurations of the planet binary (look at build.sh for some other options mostly useful for development). If you have the dependencies above installed, this should "just work".

Usage

You'll need to fetch the benchmark files from http://tampbenchmark.aass.oru.se/index.php?title=Main_Page. Take a look at the lua directory for some example predicate semantics implementations, and replace the PDDL files with those in the test_scenarios directory.

In general: run ./planet <YOUR PROBLEM CONFIG.toml>. Look at the .toml files in this repo for examples.

You can use ./planet --help to see flags to control execution. In particular, -g/--greedy uses best-first action selection (rather than prioritized sampling), which can lead to speedups for some problems.

Helpful tips for running on your own problems

• You must annotate discrete predicates as :discrete, and kinematics-altering predicates (see paper for explanation) as :kinematic.
• We use LuaJit 2.0.5, which is compatible with Lua 5.1
• Your parameters to actions and predicates (and, correspondingly, their usage in preconditions and effects for actions) must be valid Lua variable names.
• Continuous (i.e. non-discrete & non-kinematic) predicates are assumed to have a Lua function definition with the same name as the predicate in your domain semantics file.
• Continuous predicate definitions can only (as of now) use the operator functions eq, And, Or, le, ge, lt, gt, and basic math operators (note that if a math operator has an overloaded operator function you need to use the overloaded version)

Status

Planet is usable as a proof of concept. No claims are made about the quality of engineering, performance, etc. Use at your own risk.

TODO

• Remove need for sci-lua/generally simplify Lua dependencies.
• Implement syntactic transformation for handling negation
• Profiling/rewrites for performance improvements