Only released in EOL distros:
Package Summary
The descartes_planner package
- Maintainer status: developed
- Maintainer: Jorge Nicho <jnicho AT swri DOT org>, Jonathan Meyer <jonathan.meyer AT swri DOT org>, Shaun Edwards <sedwards AT swri DOT org>
- Author: Jorge Nicho <jnicho AT swri DOT org>
- License: Apache2.0
- Source: git https://github.com/ros-industrial-consortium/descartes.git (branch: hydro-devel)
Package Summary
The descartes_planner package
- Maintainer: Jorge Nicho <jnicho AT swri DOT org>, Jonathan Meyer <jonathan.meyer AT swri DOT org>, Shaun Edwards <sedwards AT swri DOT org>
- Author: Jorge Nicho <jnicho AT swri DOT org>
- License: Apache2.0
- Source: git https://github.com/ros-industrial-consortium/descartes.git (branch: indigo-devel)
Contents
Concepts
This package contains implementations of Descartes planners. For more information on the architecture of Descartes, see this page. A Descartes planner is responsible for intelligently searching the valid robot configurations for a given trajectory (with tolerances) and generating a solution in the form of a sequence of joint configurations.
Reference Implementations
Descartes currently offers two reference implementations of trajectory planners. Both accept a sequence of arbitrary trajectory points for input and output a sequence of joint trajectory points. See more details at the descartes_trajectory package page.
Dense Planner
The first and most straightforward planner is DensePlanner. It takes dense trajectories and attempts to find a path through the points that minimizes total joint motion.
Conceptually, it performs the following operations:
- For every trajectory point, calculate all of the inverse kinematic solutions for that point. These are calculated using the sampling logic implemented by the trajectory point, and the robot model's knowledge of statics/dynamics.
- For every pair of contiguous points, calculate the movement cost and feasibility between each joint solution for the starting point and each joint solution for the ending point.
- Build a graph from the data where vertexes represent kinematic solutions and edges represent movement costs.
- Search this graph for the lowest cost route from any starting configuration to any ending configuration using Dijkstra's algorithm.
Sparse Planner
Sparse Planner builds on top of the working principles of dense planner, but attempts to save computational time by 'sampling' a given trajectory. Sparse planner runs inverse kinematics on a subset of the original points and then attempts to solve the rest of the trajectory using joint space interpolation and forward kinematics.
For example, if the planner must solve for points A, B, and C. It might run inverse kinematics on A and C, and then try to infer the joint solution for B from those. It would check its inference by running forward kinematics and comparing it to the desired pose.
Limitations and Failure Modes
Currently neither planner makes any effort to resolve discontinuities in the joint space graphs generated from user trajectories and will fail if any of the following occur:
- A point in your trajectory is not reachable by the robot model.
- A point is reachable but involves a configuration flip exceeding some threshold.