This page lists potential projects that would greatly benefit the MoveIt project. This is by no means an exhaustive list, but is meant to serve as a discussion starting point for code sprints and other efforts. Please contact PickNik Robotics for further information or sponsorship opportunities.
Detailed description: TrajOpt is a motion planning framework from UC Berkeley for generating robot trajectories by local optimization. Integration of TrajOpt into the MoveIt framework would require API changes to how MoveIt stores and accesses information about the kinematic chain and changes to the MoveIt collision checking API to support integration with the Bullet collision checker for convex to convex contact checking.
Some initial work was performed by Omid Heidari during an internship at PickNik in 2019, but more work is needed to cleanly integrate the framework with MoveIt. The collision checking interface for MoveIt has changed significantly. Implementing a collision cost using this unified interface for collision checking backends is one of the major remaining issues.
MoveIt also contains implementations of other trajectory optimization methods: STOMP and CHOMP. These implementations all contain their own way of computing trajectories, but share a lot of common ideas, such as using a signed distance field to compute collision cost. Refactoring this code to use common code for the same ideas would help in making the code more maintainable. It is also a first step to making it possible to provide callback hooks for a user-configurable trajectory cost function that can be used with any of the trajectory optimization methods, which is currently not possible.
Detailed description: The idea is to identify a number of standard tasks at varying levels of difficulty that can be achieved with several robots for which a MoveIt configuration is available. Tasks can range from simply reaching for a particular goal position in a relatively uncluttered space to picking and placing items from cluttered shelves.
Once a standard benchmark has been established, subsequent work can focus on running the benchmark suite as part of a continuous integration suite. Additional work can be done on tuning hyperparameter settings that will lead to better overall performance.
Detailed description: The default cartesian planner in MoveIt is naive and has enormous room for improvement. Improvements to the Cartesian planner could include:
Detailed description: The planning scene in MoveIt is currently represented as a flat list of transforms from the origin. In this project, the planning scene should be extended with a scene graph or kinematic tree, so that it allows e.g. the dynamic attachment or removal of a hand to a robot arm, or picking up a board with other items attached to it.
Detailed description: MoveIt supports storing Constraints, PlanningScenes and RobotStates in a database, termed “warehouse” within MoveIt. This interface was recently changed to support plugins, but currently there are no plugins available that compile/work with a standard linux distribution out of the box.
Detailed description: Currently, MoveIt supports the OMPL planning library via a C++ based binding system in the package
ompl_interface. This should be deprecated in favor of a plugin-based system, that eases configuration of different planners within OMPL. In effect, the plugin allows new planning algorithms derived from the abstract Planner API in OMPL to be loaded via a yaml file. It also allows users to easily create new planning configurations for existing planning algorithms in OMPL via such a yaml. Most of the work for this is already done and has been used extensively in the latest versions. This is a great project for someone who wants to better understand the connection between MoveIt And OMPL.