Robotics and Biology Laboratory

Motion Generation

Motion Generation is concerned with the planning and execution of motion tasks while avoiding collisions. We are especially interested in motion generation for mobile manipulators operating in the real world.


Mobile manipulation takes place in unstructured environments which still poses a challenge for motion generation method. Detailed and reliable models of unstructured environments are usually not available. The robot has to perceive the world with its on-board sensors to obtain the information needed for collision-free motion and successful execution of manipulation tasks. To do so efficiently, it needs to 1) exploit rich feedback from different sensor modalities, 2) reason efficiently about the uncertainty present in the environment, and 3) balance exploration with exploitation.

Multi-modal Control for Manipulation

Contact Person

Arne Sieverling

We want to find methods for motion generation that leverage the full perceptual capabilities of mobile manipulators. These methods should integrate visual and force perception to perceive the world while moving in it. Current motion planning methods do not incorporate the high-dimensional information of currently available sensors and thus are limited in robustness. To overcome this, we shift the boundary between planning and control. Giving more responsibility to control allows us to leverage the inherent capability of controllers; dealing with dynamic situations. 

Visual servo control defines tight perception-action loops on task-relevant visual features. Force control provides complex manipulation skills using force feedback. We want to find these loops and use them for motion generation. We believe that using recursive estimation, a robot can learn multi-modal servo controllers while manipulating the world. These controllers adapt to the available information, the task, and thus provide more general and robust motion skills.

 

Motion planning

Contact Person

Arne Sieverling

Motion Planning is the task of finding a way to move a robot from one position to another while avoiding obstacles. To do this, we need to find a trajectory in the space of all possible robot configurations. For a robot with several joints, this search space is high dimensional and very complex. Even for simple two-dimensional spaces, general motion planning is NP-hard.

We think that a successful and fast motion planning algorithm has to balance between two general planning strategies: Exploration and exploitation. Exploration tries to obtain information about the connectivity of the space without considering any specific goal. Exploitation tries to find a valid path to a specific goal, using the available information.

In our motion planners, we use several sources of information to balance exploration and exploitation. Utility-guided sampling apply guided exploration by choosing samples with maximal expected utility. This approach gives us a good understanding of connectivity using much less samples than other approaches. Disassembly planning uses the 3D-workspace connectivity to identify the regions of the configuration space where a detailed search is needed. This connectivity can be obtained by expanding the workspace with a tree of bubbles. The same workspace information is used in our Exploring/Exploiting Tree. In easy regions we use the workspace information as a navigation function to drag the robot along the spheres. If this approach fails in hard regions, we gradually shift to exploration.

 

Motion Generation under Uncertainty

Contact Person

Arne Sieverling
Előd Páll

A mobile manipulator in unstructured environments needs to account for uncertainty due to noisy sensors with limited range and partially unknown, dynamic environments. General algorithms that plan under uncertainty are infeasible to use in real applications due to their high complexity. Feedback controllers provide many desirable qualities for planning under uncertainty. Controllers continuously integrate new sensor data using task-relevant modalities. These local controllers reduce uncertainty about the robots state in the environment implicitly, so much that we do not consider it during planning. A source of uncertainty we cannot ignore is the environment itself. We propose a time-dependent probabilistical model for the uncertain state of the world. In the case of an environment composed of static and moving obstacles, we can solve planning problems in polynomial time by solving an Expected Shortest Path problem. The solution to this problem is a motion policy for the robot that incorporates goal-directed sensing actions and task-consistent motion.

 

Founding

Alexander von Humboldt professorship - awarded by the Alexander von Humboldt foundation and funded through the Ministry of Education and Research, BMBF, July 2009 - June 2014

Soft Manipulation (SoMa) -  funded by the European Commission in the Horizon 2020 program, award number 645599, May 2015 - April 2019.

Flexible Skill Acquisitionen and Intuitive Robot Tasking for Mobile Manipulation in the Real World (First MM) - funded by European Commision, in the program Cognitive Systems and Robotics, award number FP7-ICT-248258, February 2010 - July 2013

Publications

2018

Pall, Elod; Sieverling, Arne; Brock, Oliver
Contingent Contact-Based Motion Planning
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Seite 6615–6621
Oktober 2018

2017

Putz, Ines; Brock, Oliver
Elastic network model of learned maintained contacts to predict protein motion
PLOS ONE, 12 (8) :e0183889
August 2017
Abele, Jessica; Brock, Oliver
First Analysis of Environment Design for Motion Planning with Contact
Juli 2017
Pall, Elod; Sieverling, Arne; Brock, Oliver
Towards Motion Plans That React to Contact Events
RSS workshop: Revisiting Contact - Turning a problem into a solution
Juli 2017
Sieverling, Arne; Eppner, Clemens; Brock, Oliver
Exploiting Contact for Efficient Motion Planning Under Uncertainty
RSS workshop: Revisiting Contact - Turning a problem into a solution
2017
Sieverling, Arne; Eppner, Clemens; Wolff, Felix; Brock, Oliver
Interleaving Motion in Contact and in Free Space for Planning Under Uncertainty
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Seite 4011-4017
2017

2015

Lehner, Peter; Sieverling, Arne; Brock, Oliver
Incremental, Sensor-Based Motion Generation for Mobile Manipulators in Unknown, Dynamic Environments
Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), Seite 4761-4767
Mai 2015

2014

Rickert, Markus; Sieverling, Arne; Brock, Oliver
Balancing Exploration and Exploitation in Sampling-Based Motion Planning
IEEE Transactions on Robotics, 30 (6) :1305-1317
Dezember 2014
Sieverling, Arne; Kuhnen, Nicolas; Brock, Oliver
Sensor-Based, Task-Constrained Motion Generation Under Uncertainty
Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), Seite 4348-4355
Mai 2014

2011

Rickert, Markus
Efficient Motion Planning for Intuitive Task Execution in Modular Manipulation Systems
Technische Universität München, Munich, Germany
2011

2010

Kamm, Florian
Using Tree-Based Robot Motion Planning Algorithms for Protein Loop Closure
Institute of Biology, Humboldt University of Berlin, Germany
Dezember 2010
Yang, Yuandong; Brock, Oliver
Elastic Roadmaps - Motion Generation for Autonomous Mobile Manipulation
28 (1) :113-130
2010
ISSN: 0929-5593(Print) 1573-7527 (Online)

2008

Rickert, Markus; Brock, Oliver; Knoll, Alois
Balancing Exploration and Exploitation in Motion Planning
IEEE International Conference on Robotics and Automation, Seite 2812-2817
Herausgeber: Pasadena, USA
Mai 2008
ISBN
978-1-4244-1647-9

2007

Burns, Brendan; Brock, Oliver
Sampling-Based Motion Planning with Sensing Uncertainty
IEEE International Conference on Robotics and Automation, Seite 3313-3318
Herausgeber: Rome, Italy
April 2007
ISBN
1-4244-0602-1/07
Burns, Brendan; Brock, Oliver
Single-Query Motion Planning with Utility-Guided Random Trees
Proceedings of the International Conference on Advanced Robotics, Seite 3307-3312
Herausgeber: Rome, Italy
April 2007
Burns, Brendan; Brock, Oliver
Single-Query Motion Planning with Utility-Guided Random Trees
In Proceedings of the IEEE International Conference on Robotics and Automation, Seite 3307-3312
Herausgeber: Rome, Italy
April 2007

2006

Burns, Brendan; Brock, Oliver
Sampling-Based Motion Planning Using Uncertain Knowledge
Department of Computer Science, University of Massachusetts Amherst
Technical Report
Juni 2006
Burns, Brendan; Brock, Oliver
Utility-Guided Random Trees
aus Technical Report
Department of Computer Science, University of Massachusetts Amherst
Juni 2006

2005

Yang, Yuandong; Brock, Oliver
Efficient Motion Planning Based on Disassembly
Proceedings of Robotics: Science and Systems, Cambridge, Seite 01-08
Herausgeber: USA
Juni 2005
Burns, Brendan; Brock, Oliver
Toward Optimal Configuration Space Sampling
Proceedings of Robotics: Science and Systems, Seite 105-112
Herausgeber: Cambridge, USA
Juni 2005
Burns, Brendan; Brock, Oliver
Single-Query Entropy-Guided Path Planning
Proceedings of the International Conference on Advanced Robotics, Seite 2124-2129
Herausgeber: Barcelona, Spain
April 2005
ISBN
0-7803-8914-X/05
Burns, Brendan; Brock, Oliver
Sampling-Based Motion Planning Using Predictive Models
Proceedings of the International Conference on Advanced Robotics, Seite 3120-3125
Herausgeber: Barcelona, Spain
April 2005
ISBN
0-7803-8914-X/05
Burns, Brendan; Brock, Oliver
Motion Planning based on Utility Theory
Unpublished manuscript
2005

2004

Yang, Yuandong; Brock, Oliver
Viewing Motion Planning as Disassembly: A Decomposition-Based Approach for Non-Stationary Robots
Department of Computer Science, University of Massachusetts Amherst
Technical Report
Dezember 2004
Burns, Brendan; Brock, Oliver
Model-Based Motion Planning
aus Technical Report
Department of Computer Science, University of Massachusetts Amherst, Department of Computer Science, University of Massachusetts Amherst
2004

Seite 1 von 2