Machine Learning for Kinodynamic Planning (ML4KP)
Edgar Granados*, Aravind Sivaramakrishnan*, Troy McMahon, Zakary Littlefield,
Kostas E. Bekris
PRACSYS Lab @ Rutgers University
Machine Learning for Motion Planning (MLMP) Workshop @ ICRA 2021
Edgar Granados*, Aravind Sivaramakrishnan*, Troy McMahon, Zakary Littlefield,
Kostas E. Bekris
PRACSYS Lab @ Rutgers University
Machine Learning for Motion Planning (MLMP) Workshop @ ICRA 2021
We introduce ML4KP, a C++ library for efficient kinodynamic motion planning. The library is fast, with minimal dependencies and contains implementations of state-of-the-art kinodynamic planners that can be used with minimal configuration for the included systems. By using a functional approach, the library makes it easy to integrate machine learning methods into the planning process without dealing with the core implementation of the planner. This is primarily achieved by defining the specification of each planner to use parameters as well as functions (with default implementations) that can be bound to other user-defined implementations. In contrast to other motion planning libraries, ML4KP allows to directly generate solutions for dynamic systems, with optional interfaces to physics simulators as well as Python bindings. ML4KP is intended to be used by the robot learning community that seek to use sampling-based motion planners, as well as the motion planning community that wants to integrate machine learning tools.
Link to technical report: [pdf]
Link to code: https://github.com/PRX-Kinodynamic/ML4KP
@MISC{ML4KP,
author = {Edgar Granados and Aravind Sivaramakrishnan and Troy McMahon and Zakary Littlefield and Kostas E. Bekris},
title = {Machine Learning for Kinodynamic Planning (ML4KP)},
howpublished = {\url{https://github.com/PRX-Kinodynamic/ML4KP}},
year = {2021--2021}
}