Publications

@InProceedings{osswald18iros,
  Title      = {{GPU}-Accelerated Next-Best-View Coverage of Articulated Scenes},
  Author     = {Stefan O{\ss}wald and Maren Bennewitz},
  Booktitle  = {Proc.\ of the {IEEE/RSJ} International Conference on Intelligent Robots and Systems (IROS)},
  Year       = {2018},
  Abstract   = {Next-best-view algorithms are commonly used for covering known scenes, for example in search, maintenance, and mapping tasks. In this paper, we consider the problem of planning a strategy for covering articulated environments where the robot also has to manipulate objects to inspect obstructed areas. This problem is particularly challenging due to the many degrees of freedom resulting from the articulation. We propose to exploit graphics processing units present in many embedded devices to parallelize the computations of a greedy next-best-view approach. We implemented algorithms for costmap computation, path planning, as well as simulation and evaluation of viewpoint candidates in OpenGL for Embedded Systems and benchmarked the implementations on multiple device classes ranging from smartphones to multi-GPU servers. We introduce a heuristic for estimating a utility map from images rendered with strategically placed spherical cameras and show in simulation experiments that robots can successfully explore complex articulated scenes with our system.}
}

@InProceedings{sheikh18iros,
  Title      = {A Combined {RGB} and Depth Descriptor for {SLAM} with Humanoids},
  Author     = {Rasha Sheikh and Stefan O{\ss}wald and Maren Bennewitz},
  Booktitle  = {Proc.\ of the {IEEE/RSJ} International Conference on Intelligent Robots and Systems (IROS)},
  Year       = {2018},
  Abstract   = {In this paper, we present a visual simultaneous localization and mapping (SLAM) system for humanoid robots.  We introduce a new binary descriptor called DLab that exploits the combined information of color, depth, and intensity to achieve robustness with respect to uniqueness, reproducibility, and stability.  We use DLab within ORB-SLAM, where we replaced the place recognition module with a modification of FAB-MAP that works with newly built codebooks using our binary descriptor.  In experiments carried out in simulation and with a real Nao humanoid equipped with an RGB-D camera, we show that DLab has a superior performance in comparison to other descriptors. The application to feature tracking and place recognition reveal that the new descriptor is able to reliably track features even in sequences with seriously blurred images and that it has a higher percentage of correctly identified similar images.  As a result, our new visual SLAM system has a lower absolute trajectory error in comparison to ORB-SLAM and is able to accurately track the robot's trajectory.}
}

@InProceedings{osswald17humanoids,
  Title      = {Efficient Coverage of {3D} Environments with Humanoid Robots Using Inverse Reachability Maps},
  Author     = {Stefan O{\ss}wald and Philipp Karkowski and Maren Bennewitz},
  Booktitle  = {Proc.\ of the {IEEE-RAS} International Conference on Humanoid Robots (HUMANOIDS)},
  Year       = {2017},
  Doi        = {10.1109/HUMANOIDS.2017.8239550},
  Pages      = {151--157},
  Abstract   = {Covering a known 3D environment with a robot's camera is a commonly
  required task, for example in inspection and surveillance, mapping, or
  object search applications.  In addition to the problem of finding a
  complete and efficient set of view points for covering the whole
  environment, humanoid robots also need to observe balance, energy, and
  kinematic constraints for reaching the desired view poses.  In this
  paper, we approach this high-dimensional planning problem by introducing
  a novel inverse reachability map representation that can be used for
  fast pose generation and combine it with a next-best-view algorithm for
  covering a known 3D environment.  We implemented our approach in ROS and
  tested it with a Nao robot on both simulated and real-world scenes. The
  experiments show that our approach enables the humanoid to efficiently
  cover room-sized environments with its camera.}
}

@InProceedings{karkowski16humanoids,
  author    = {Philipp Karkowski and Stefan O{\ss}wald and Maren Bennewitz},
  title     = {Real-Time Footstep Planning in {3D} Environments},
  booktitle = {Proc.\ of the {IEEE-RAS} Int.\ Conf.\ on Humanoid Robots (HUMANOIDS)},
  year      = {2016},
  pages     = {69--74},
  doi       = {10.1109/HUMANOIDS.2016.7803256},
  abstract  = {A variety of approaches exist that tackle the problem of humanoid
  locomotion. The spectrum ranges from dynamic walking controllers that
  allow fast walking to systems that plan longer footstep paths through
  complicated scenes. Simple walking controllers do not guarantee
  collision-free steps, whereas most existing footstep planners are not
  capable of providing results in real time. Thus, these methods cannot be
  used, not even in combination, to react to sudden changes in the
  environment. In this paper, we propose a new fast search method that
  combines A* with an adaptive 3D action set. When expanding a node, we
  systematically search for suitable footsteps by taking into account
  height information. As we show in various experiments, our approach
  outperforms standard A*-based footstep planning in both run time and
  path cost and, combined with an efficient map segmentation, finds valid
  footstep plans in 3D environments in under 50 ms.}
}

@InProceedings{regier16iros,
  author    = {Peter Regier and Stefan O{\ss}wald and Philipp Karkowski and Maren Bennewitz},
  title     = {Foresighted Navigation Through Cluttered Environments},
  booktitle = {Proc.\ of the {IEEE/RSJ} Int.\ Conf.\ on Intelligent Robots and Systems (IROS)},
  year      = {2016},
  pages     = {1437--1442},
  doi       = {10.1109/IROS.2016.7759234},
  abstract  = {In this paper, we introduce an approach to efficient robot navigation
  through cluttered indoor environments. We propose to estimate local
  obstacle densities based on already detected objects and use them to
  predict traversal costs corresponding to potential obstacles in regions
  not yet observable by the robot's sensors. By taking into account the
  predicted costs for path planning, the robot is then able to navigate in
  a more foresighted manner and reduces the risk of getting stuck in
  cluttered regions. We thoroughly evaluated our approach in simulated
  and real-world experiments. As the experimental results demonstrate, our
  method enables the robot to efficiently navigate through environments
  containing cluttered regions and achieves significantly shorter
  completion times compared to a standard approach not using any
  prediction.}
}

@Article{osswald16ral,
  author    = {Stefan O{\ss}wald and Maren Bennewitz and Wolfram Burgard and Cyrill Stachniss},
  title     = {Speeding-Up Robot Exploration by Exploiting Background Information},
  journal   = {{IEEE} Robotics and Automation Letters (RA-L)},
  year      = {2016},
  volume    = {1},
  number    = {2},
  pages     = {716--723},
  doi       = {10.1109/LRA.2016.2520560},
  issn      = {2377-3766},
  abstract  = {The ability to autonomously learn a model of an environment 
  is an important capability of a mobile robot. In this paper, we 
  investigate the problem of exploring a scene given background information 
  in form of a topo-metric graph of the environment.  Our method is 
  relevant for several real-world applications in which the rough structure
  of the environment is known beforehand. We present an approach that 
  exploits such background information and enables a robot to cover the 
  environment with its sensors faster compared to a greedy exploration 
  system without this information.  We implemented our exploration system 
  in ROS and evaluated it in different environments. As the experimental 
  results demonstrate, our proposed method significantly reduces the 
  overall trajectory length needed to cover the environment with the 
  robot's sensors and thus yields a more efficient exploration strategy 
  compared to state-of-the-art greedy exploration, if the additional 
  information is available.}
}

@inproceedings{osswald14icra,
  author    = {Stefan O{\ss}wald and Henrik Kretzschmar and Wolfram Burgard and Cyrill Stachniss},
  title     = {Learning to Give Route Directions from Human Demonstrations},
  booktitle = {Proc.\ of the {IEEE} International Conference on Robotics \& Automation (ICRA)},
  year      = {2014},
  address   = {Hong Kong, China},
  doi       = {10.1109/ICRA.2014.6907334},
  pages     = {3303-3308}, 
  url       = {http://ais.informatik.uni-freiburg.de/publications/papers/osswald14icra.pdf},
  abstract  = {For several applications, robots and other computer systems 
  must provide route descriptions to humans. These descriptions should be 
  natural and intuitive for the human users. In this paper, we present an 
  algorithm that learns how to provide good route descriptions from a 
  corpus of human-written directions. Using inverse reinforcement learning, 
  our algorithm learns how to select the information for the description 
  depending on the context of the route segment. The algorithm then uses 
  the learned policy to generate directions that imitate the style of the 
  descriptions provided by humans, thus taking into account personal as 
  well as cultural preferences and special requirements of the particular 
  user group providing the learning demonstrations. We evaluate our 
  approach in a user study and show that the directions generated by our 
  policy sound similar to human-given directions and substantially more 
  natural than directions provided by commercial web services.},
}

@article{hornung14ijhr,
  AUTHOR = {Armin Hornung and Stefan O{\ss}wald and Daniel Maier and Maren Bennewitz},
  TITLE = {Monte {C}arlo Localization for Humanoid Robot Navigation in Complex Indoor Environments},
  YEAR = 2014,
  JOURNAL = {International Journal of Humanoid Robotics (IJHR)},
  DOI = {10.1142/S0219843614410023},
  volume = {11},
  number = {2}
}

@InProceedings{osswald12iros,
  author =       {Stefan O{\ss}wald and Armin Hornung and Maren Bennewitz},
  title =        {Improved Proposals for Highly Accurate Localization Using Range and Vision Data},
  booktitle =    {Proc.\ of the {IEEE/RSJ} International Conference on Intelligent Robots and Systems (IROS)},
  year =         {2012},
  pages =        {1809--1814},
  month =        oct,
  doi =          {10.1109/IROS.2012.6385657},
  issn =         {2153-0858},
  url =          {http://hrl.informatik.uni-freiburg.de/papers/osswald12iros.pdf},
  abstract =     {In order to successfully climb challenging stair-cases that 
  consist of many steps and contain difficult parts, humanoid robots need to 
  accurately determine their pose. In this paper, we present an approach that 
  fuses the robot's observations from a 2D laser scanner, a monocular camera, an 
  inertial measurement unit, and joint encoders in order to localize the robot 
  within a given 3D model of the environment. We develop an extension to 
  standard Monte Carlo localization (MCL) that draws particles from an improved 
  proposal distribution to obtain highly accurate pose estimates. Furthermore, 
  we introduce a new observation model based on chamfer matching between edges 
  in camera images and the environment model. We thoroughly evaluate our 
  localization approach and compare it to previous techniques in real-world 
  experiments with a Nao humanoid. The results show that our approach 
  significantly improves the localization accuracy and leads to a considerably
  more robust robot behavior. Our improved proposal in combination with chamfer 
  matching can be generally applied to improve a range-based pose estimate by a 
  consistent matching of lines obtained from vision.
  }
}

@INPROCEEDINGS{osswald12sc,
  author = {Stefan O{\ss}wald and Armin Hornung and Maren Bennewitz},
  title = {Accurate {6D} Localization in Multi-Level Environments},
  booktitle = {Extended Abstracts of Spatial Cognition (SC)},
  url = {http://sc2012.informatik.uni-freiburg.de/posters/sc20120extendedabstract-30.pdf},
  year = {2012},
  month = aug
}

@MASTERSTHESIS{osswald12msc,
  author = {Stefan O{\ss}wald},
  title = {Techniques for Autonomous Stair Climbing with Humanoid Robots},
  school = {University of Freiburg, Department of Computer Science, Humanoid
	Robots Lab},
  year = {2012},
  type = {Master's Thesis},
  month = feb,
  abstract = {Service robots need to be able to reliably climb stairs in order to
	act autonomously in indoor environments. In this thesis, we present
	techniques that enable a Nao humanoid robot equipped with a laser
	range finder to autonomously climb up a spiral staircase in a complex
	multi-level environment. In contrast to other approaches, we use
	a standard platform robot without external sensors or specialized
	hardware components for the stair climbing task, and we do not modify
	the environment in order to help the robot to sense the stairs.
	
	
	In order to climb up multiple steps of a complex staircase in a row,
	the robot needs to accurately determine its pose on the stairs. In
	this thesis, we first discuss and evaluate a standard Monte Carlo
	Localization (MCL) approach that fuses observations from the 2D laser
	scanner, an inertial measurement unit, and joint encoders in order
	to localize the robot within a given 3D model of the environment.
	We then present two extensions to this approach that increase the
	accuracy of the pose estimate by additionally integrating information
	from images acquired with an on-board camera. The first extension
	reconstructs a 3D model of the next step from edges observed in the
	camera images and estimates the robot's pose relative to the reconstructed
	model. The second approach draws particles from an improved proposal
	distribution, which yields a better representation of the robot's
	pose posterior distribution. Additionally, the second extension introduces
	an observation model based on chamfer matching that fits an edge
	model of the staircase consistently to a set of images.
	
	
	We thoroughly evaluate and compare our localization approaches in
	real-world experiments. The results show that our extensions to the
	standard MCL approach improve the localization accuracy significantly
	and enable the robot to accurately determine its pose. Using our
	approach, the robot can reliably climb the steps of a spiral staircase.}
}

@INPROCEEDINGS{osswald11humanoids,
  author = {Stefan O{\ss}wald and Jens-Steffen Gutmann and Armin Hornung and
	Maren Bennewitz},
  title = {From 3{D} Point Clouds to Climbing Stairs: A Comparison of Plane
	Segmentation Approaches for Humanoids},
  booktitle = {Proc.\ of the {IEEE-RAS} International Conference on Humanoid Robots
	(Humanoids)},
  year = {2011},
  pages = {93--98},
  month = oct,
  abstract = {In this paper, we consider the problem of building 3D models of complex
	staircases based on laser range data acquired with a humanoid. These
	models have to be sufficiently accurate to enable the robot to reliably
	climb up the staircase. We evaluate two state-of-the-art approaches
	to plane segmentation for humanoid navigation given 3D range data
	about the environment. The first approach initially extracts line
	segments from neighboring 2D scan lines, which are successively combined
	if they lie on the same plane. The second approach estimates the
	main directions in the environment by randomly sampling points and
	applying a clustering technique afterwards to find planes orthogonal
	to the main directions. We propose extensions for this basic approach
	to increase the robustness in complex environments which may contain
	a large number of different planes and clutter. In practical experiments,
	we thoroughly evaluate all methods using data acquired with a laser-equipped
	Nao robot in a multi-level environment. As the experimental results
	show, the reconstructed 3D models can be used to autonomously climb
	up complex staircases.},
  doi = {10.1109/Humanoids.2011.6100836},
  url = {http://hrl.informatik.uni-freiburg.de/papers/osswald11humanoids.pdf}
}

@INPROCEEDINGS{osswald11iros,
  author = {O{\ss}wald, Stefan and G\"{o}r\"{o}g, Attila and Hornung, Armin and
	Bennewitz, Maren},
  title = {Autonomous climbing of spiral staircases with humanoids},
  booktitle = {Proc.\ of the {IEEE/RSJ} International Conference on Intelligent Robots
	and Systems (IROS)},
  year = {2011},
  pages = {4844--4849},
  month = sep,
  abstract = {In this paper, we present an approach to enable a humanoid robot to
	autonomously climb up spiral staircases. This task is substantially
	more challenging than climbing straight stairs since careful repositioning
	is needed. Our system globally estimates the pose of the robot, which
	is subsequently refined by integrating visual observations. In this
	way, the robot can accurately determine its relative position with
	respect to the next step. We use a 3D model of the environment to
	project edges corresponding to stair contours into monocular camera
	images. By detecting edges in the images and associating them to
	projected model edges, the robot is able to accurately locate itself
	towards the stairs and to climb them. We present experiments carried
	out with a Nao humanoid equipped with a 2D laser range finder for
	global localization and a low-cost monocular camera for short-range
	sensing. As we show in the experiments, the robot reliably climbs
	up the steps of a spiral staircase.},
  doi = {10.1109/IROS.2011.6048209},
  issn = {2153-0858},
  keywords = {2D laser range finder;3D model;Nao humanoid;autonomous climbing;edge
	detection;global localization;humanoid robot;monocular camera images;robot
	pose estimation;spiral staircases;stair contours;edge detection;humanoid
	robots;laser ranging;mobile robots;motion control;pose estimation;position
	control;},
  url = {http://hrl.informatik.uni-freiburg.de/papers/osswald11iros.pdf}
}

@INPROCEEDINGS{osswald10icra,
  author = {Stefan O{\ss}wald and Armin Hornung and Maren Bennewitz},
  title = {Learning Reliable and Efficient Navigation with a Humanoid},
  booktitle = {Proc.\ of the {IEEE} International Conference on Robotics \& Automation
	(ICRA)},
  year = {2010},
  pages = {2375--2380},
  month = may,
  abstract = {Reliable and efficient navigation with a humanoid robot is a difficult
	task. First, the motion commands are executed rather inaccurately
	due to backlash in the joints or foot slippage. Second, the observations
	are typically highly affected by noise due to the shaking behavior
	of the robot. Thus, the localization performance is typically reduced
	while the robot moves and the uncertainty about its pose increases.
	As a result, the reliable and efficient execution of a navigation
	task cannot be ensured anymore since the robot's pose estimate might
	not correspond to the true location. In this paper, we present a
	reinforcement learning approach to select appropriate navigation
	actions for a humanoid robot equipped with a camera for localization.
	The robot learns to reach the destination reliably and as fast as
	possible, thereby choosing actions to account for motion drift and
	trading off velocity in terms of fast walking movements against accuracy
	in localization. We present extensive simulated and practical experiments
	with a humanoid robot and demonstrate that our learned policy significantly
	outperforms a hand-optimized navigation strategy.},
  doi = {10.1109/ROBOT.2010.5509420},
  url = {http://hrl.informatik.uni-freiburg.de/papers/osswald10icra.pdf}
}

@InProceedings{hornung10erlars,
  author =       {Armin Hornung and Maren Bennewitz and Cyrill Stachniss and Hauke Strasdat and Stefan O{\ss}wald and Wolfram Burgard},
  title =        {Learning Adaptive Navigation Strategies for Resource-Constrained Systems},
  booktitle =    {Proc.\ of the 3rd International Workshop on
Evolutionary and Reinforcement Learning for Autonomous Robot Systems (ERLARS)},
  year =         {2010},
  month = aug,
  pages = {1--10},
  issn = {2190-5576},
  url = {http://hrl.informatik.uni-freiburg.de/papers/hornung10erlars.pdf},
  abstract = {The majority of navigation algorithms for mobile robots
assume that the robots possess enough computational or memory
resources to carry out the necessary calculations. Especially small
and lightweight devices, however, are resource-constrained and have
only restricted capabilities. In this paper, we present a reinforcement
learning approach for mobile robots that considers the imposed constraints 
on their sensing capabilities and computational resources, so that they 
can reliably and efficiently fulfill their navigation tasks. Our
technique learns a policy that optimally trades off the speed of the
robot and the uncertainty in the observations imposed by its movements. 
It furthermore enables the robot to learn an efficient landmark selection 
strategy to compactly model the environment. We describe extensive 
simulated and real-world experiments carried out with both wheeled and 
humanoid robots which demonstrate that our learned navigation policies 
significantly outperform strategies using advanced and manually optimized 
heuristics.}
}

@MASTERSTHESIS{osswald09bsc,
  author = {Stefan O{\ss}wald},
  title = {Reliable Vision-based Navigation with a Humanoid Robot},
  school = {University of Freiburg, Department of Computer Science, Humanoid
	Robots Lab},
  year = {2009},
  type = {Bachelor's Thesis},
  month = sep,
  abstract = {Motion blur is a severe problem in visual localization since it can
	prevent a robot from detecting and matching visual features reliably.
	Especially when using a small humanoid robot, the images taken while
	the robot is walking are strongly affected by motion blur. Thus,
	the localization performance is typically reduced in these situations,
	which means that the uncertainty of the robot about its pose increases.
	As a result, it cannot be ensured anymore that a navigation task
	can be executed reliably and efficiently since the robot's pose estimate
	might not correspond to the true location.
	
	In this thesis, we extend an existing reinforcement learning approach
	for wheeled robots so that it is applicable to the domain of humanoid
	robots. The robot learns to select appropriate navigation actions
	to reach the destination reliably and as fast as possible, thereby
	trading off velocity against accuracy in localization. The robot
	implicitly takes the impact of motion blur on observations into account
	and avoids delays caused by localization errors.
	
	Simulated and real-world experiments with the humanoid robot Nao show
	that our approach generates suitable policies for accomplishing the
	navigation task. The policies learned in our simulator can be transferred
	to the real robot and significantly outperform a standard navigation
	strategy in terms of reliability and efficiency.}
}