Ruprecht-Karls-Universität Heidelberg
Optimization in Robotics and Biomechanics

Generating optimal paths for industrial and humanoid robots

Martin Felis, Wael Suleiman, Benjamin Reh, Katja Mombaur - ORB

Antonio El Khoury, Florent Lamiraux, Michel Taix - LAAS-CNRS, Toulouse

 

This project is financed by the European Union as part of the ECHORD project.

The generation of the best possible path that does not violate any constraints imposed by the environment is an ubiquitous task in both industrial and humanoid robotics.

Currently there is no algorithmic approach available that allows to address this problem for very complex dynamic robot systems in cluttered changing environments in real time.

Instead there are two established but still quite separated research areas that both address a part of the problem, namely path planning and numerical optimal control:

  • Path planning is mainly interested in the determination of a feasible path in very complex environments based on geometric and kinematic models.
  • Numerical optimal control techniques are capable to generate optimal trajectories for robot manipulators or humanoid robots taking into account the dynamics. However the treatment of a huge number of environmental constraints giving rise to many local minima makes the problems very hard, if not impossible, to solve.


In our ECHORD experiment GOP we developed a framework that combines the two methods and tested it successfully on an industrial robot arm, namely the KUKA KR5 sixx R850, and the humanoid robot HRP-2.

For this we used the following software components:

  • KiteLab [1], a framework for randomized motion planners
  • hpp-constrained [2], a package built on top of KiteLab to allow path planning on constraint manifolds
  • MUSCOD-II [3], a software package for optimal control problems
  • RBDL [4], the Rigid Body Dynamics Library was used for the modeling of the KUKA KR5 sixx R850

GOP Framework:

In a first step, a draft geometric planner (KiteLab) finds a collision-free quasi-static path, given a robot and a set of obstacles. This planner is
guaranteed to find a solution if it exists. This draft path is then locally reshaped and time-parametrized by a numerical optimal control solver
(MUSCOD-II) to produce a collision-free optimal and feasible trajectory.

Scenarios:

  • KUKA KR sixx R850: Two scenarios were created for the robot arm KUKA KR sixx R850 which would move a glass that contained liquid in a complex environment without spilling any of the liquid. The scenarios were designed so that the the robot has to move in a non-trivial way to move from the starting position to the goal position without spilling liquid.
  • HRP-2: In this experimental setup the humanoid robot HRP-2 has to pick an object inside a shelf and displace it to an upper level while avoiding collisions with itself and the environment.

 

Achievements:


During the GOP-ECHORD experiment we established a framework that combines Path Planning methods with Numerical Optimal Control methods. The framework was implemented in software using the Path Planning software package Kitelab and the Numerical Optimal Control Code MUSCOD-II and verified using real robots. The implementation is in general independent of the robot and the specific task and can thus be used for a wide range of robots and scenarios. The paths we obtained using our combined approach are both collision-free and dynamic in a sense that it would be impossible to obtain using just either Path Planning or Numerical Optimal Control. Also the successful transfer of the obtained motions to the robots support the practical use of this combined approach.

Multimedia report about ECHORD GOP results

Click here to open the video on youtube


[1] www.kineocam.com/kitelab/
[2] github.com/laas/hpp-constrained-planner
[3] www.iwr.uni-heidelberg.de/~agbock/RESEARCH/muscod.php
[4] rbdl.bitbucket.org

 

 

 

 

 

 

K. Mombaur, orb@uni-hd.de
Last Update: 27.06.2014 - 17:42

The photographs in the header of this webpage have been taken at the Musee de l'Automate in Souillac, France