Skip to content

A move_base ROS global_planner plug-in that quickly finds from a socially-informed Voronoi diagram a set of homotopy classes and generates a kinodynamic trajectory, into the best class, by using an optimal sampling-based motion planner

Notifications You must be signed in to change notification settings

srl-freiburg/srl_rhcf_planner

Repository files navigation

SRL_RHCF_PLANNER

Randomized Homotopy Classes Finder for Social Navigation developed within the context of the EU FP7 project SPENCER.

alt tag

Description of the figure: The red socially-informed Voronoi diagram, which implicitly encodes homotopy classes, describes the possible ways to go through a crowd in a room. Our approach rapidly selected two possible paths (in yellow and green).

Motivation

To deal with unexpected obstacles and quickly react to dynamic world's changes in social settings we utilize a fast random walk approach (RHCF) to generate a set of K distinct paths, belonging to different homotopy classes. The best geometric path among those found by RHCF is then used to generate a kinodynamic trajectory by using an optimal sampling based motion planner.

Description of the Package

SRL_RHCF_PLANNER is a ROS move_base package that implements a fast hierarchical motion planning framework, which reactively generates an optimal kinodynamic trajectory in the best homotopy class found by the algorithm Randomized Homotopy Class Finder (RHCF).

The package is a global_planner plug-in for move_base. It adhers to the specifics of nav_core::BaseGlobalPlanner, please check for further details on move_base refer to http://wiki.ros.org/move_base.

In this global planner, firstly a set of homotopy classes is generated using RHCF from a socially-informed Voronoi diagram, then a nonholonomic RRT* based algorithm generates a smooth kinodynamic trajectory in the best homotopy class among those found by RHCF.

alt tag

Requirements

  • ROS (including visualization rools -> rviz), tested on Indigo and Hydro
  • ros-hydro-navigation or ros-indigo-navigation
  • Eigen3
  • Boost >= 1.46
  • C++11 compiler
  • spencer_people_tracking package (state-of-the-art people tracker) from the EU-Project Spencer github repository, https://github.com/spencer-project/spencer_people_tracking

Installation

Clone the package into you catkin workspace

  • cd [workspace]/src
  • git clone https://github.com/srl-freiburg/srl_rhcf_planner.git
  • cd ../
  • catkin_make or catkin build

Usage

  • roslaunch srl_rhcf_planner move_base_global_srl.launch will launch the global planner node. You can launch the planner with different configurations, by varying some parameters:
    • useVoronoiRandWalk set to true if want to use the random walk, otherwise other planning methods could be choosed (see afterwards);
    • type_weight, defined the Distance used to generate the Voronoi Diagram:
      • 0, Euclidean Distance;
      • 1, Socially informed Distance;
    • weighting_gain, gain associated to the humans direction, standard value equal to 10
    • social_gain, gain associated to the social force in the scene, standard value equal to 2
    • border_gain, gain used to push away the voronoi edges from the humans, standard value equal to 7
    • discounting_factor, used to tune the speed of the random walk. The number of random walks needed to generate K distinct paths, and consequently RHCF runtime, decreases monotonically as α goes from 1 to 0.5. Standard value equal to 0.8
    • K_homotopy_classes, number of homotopy classes to generates, among those select the best one and
    • SHOW_LOG_INFO, if set to true extra info on the random walk are displayed
    • TYPE_PLANNER, set to:
      • 0, use RRT
      • 1, use RRT* only partial rewiring
      • 2, use RRT*
    • NUMBER_UPDATE_TRAJ, set to:
      • Choose after how many cost improvements the planner could stop, currently set at 2. Minimum value is 1. Higher the value, higher the computaion time required to generate a trajectory
    • BOX :
      • if it is set to 1, the nearest vertex is selected from a weighted box according to the Ball-Box Theorem.
    • RADIUS :
      • the size of the radius where the near neighbor set is generated, in case you use RRT* select -1 so to have the RRT* shrinking ball.
    • RHO :
      • end condition for the POSQ steer function, should be set to a value of few cm.
    • DT :
      • integration time step of the POSQ steer function, maximum value 0.5s
    • TYPE_SAMPLING :
      • if TYPE_SAMPLING == 0 support set as uniform over a strips following a discrete path generate by a Theta* algorithm
      • If TYPE_SAMPLING == 1 support set as Gaussian Mixture over the Theta* path
      • if TYPE_SAMPLING == 2 support set as gaussians over a spline fitting the Theta* waypoints
      • if TYPE_SAMPLING == 3 support for Theta*-RRT, if set need to specify the range where to set orientations OR_RANGE and the width of the strip along the Theta* path WIDTH_STRIP
      • if TYPE_SAMPLING == 4 support set as the entire state space, the dimension of the state space are read from the grid generate by the move_base framework
      • if TYPE_SAMPLING == 5 Path Biasing along the current available trajectory. If used need to set also the biasing probability BIAS_PROB and the DISPERSION
    • GOAL_BIASING
      • if set to 1 activate goal biasing.
    • GOAL_BIASING_THS
      • set the probability to select a state not in the state space
    • ADD_COST_FROM_COSTMAP, set to true if you want to add cost from global cost map
    • ADD_COST_PATHLENGTH, set to true if you want to add the cost associated to path length and changes of heading
    • ADD_COST_THETASTAR, set to true if you want to add cost resembling closeness to thetastar path
    • Params related to the distance metric, only one of them shoul be set to 1. LEARNED and NOTLEARNED select the best vertex from the spherical neighborhood with radius equal to the parameter RADIUS:
      • LEARNED, set to 1, if you want to find the nearest vertex according to the learned cost
      • FINDNEAREST, set to 1 if you want to find the nearest vertex according to the Kd Tree Euclidean Distance
      • NOTLEARNED, set to 1 if you want to find the nearest vertex according to the cost computed over extensions of POSQ path
    • TIMECOUNTER, set to 1 if you want to specify the maximum amount of seconds your planner should work.
    • MAXTIME, max number of seconds allowed to find a path
    • max_iterations, if TIMECOUNTER is 0, this is the maximum number of iterations the planner will execute to find a path,

Developers

Any contribution to the software is welcome. Contact the current developers for any info:

TODOs:

About

A move_base ROS global_planner plug-in that quickly finds from a socially-informed Voronoi diagram a set of homotopy classes and generates a kinodynamic trajectory, into the best class, by using an optimal sampling-based motion planner

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published