@INPROCEEDINGS{daun_large_scale_slam_2019,
  author = {Kevin Daun and Stefan Kohlbrecher and Jürgen Sturm and Oskar von Stryk},
  title = {Large Scale 2D Laser SLAM using Truncated Signed Distance Functions},
  year = {2019},
  pages = {222-228},
  booktitle = {Proceedings of the IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR)},
  keywords = {optical radar;rescue robots;SLAM (robots);tree searching;truncated signed distance functions;autonomous mobile rescue robots;grid cell;occupancy grid based methods;TSDF;unknown unstructured GPS-denied environments;lidar-based methods;consistent mapping approach;simultaneous localization and mapping approach;large scale 2D laser SLAM approach;efficient branch-and-bound based loop closure detection;urban search and rescue elements;Simultaneous localization and mapping;Real-time systems;Two dimensional displays;Optimization;Lasers;Robustness},
  doi = {10.1109/SSRR.2019.8848964},
  pdf = {2019_daun_ssrr.pdf},
  abstract = {For deployment in previously unknown, unstructured and GPS-denied environments, autonomous mobile rescue robots need to localize themselves in the environment and create a map of it using a simultaneous localization and mapping (SLAM) approach. While most existing lidar-based methods use occupancy grids to represent a map, the use of truncated signed distance functions (TSDFs) is investigated in this paper to improve accuracy and robustness. In contrast to occupancy grids, TSDFs represent the distance to the nearest surface in every grid cell. This enables sub-pixel precision during localization and increases the basin of convergence of scan matching. To enable consistent mapping of large spaces, an efficient branch-and-bound based loop closure detection is applied. The evaluation of the proposed approach with publicly available benchmark data shows that the proposed approach yields improved accuracy in comparison to occupancy grid based methods, while requiring similar runtime. Furthermore, it is demonstrated that the proposed approach is able to map a large scale environment with urban search and rescue elements in real-time.},
}

@INPROCEEDINGS{gebhardt_learning_robust_policies_2018,
  author = {Gregor H. W. Gebhardt and Kevin Daun and Marius Schnaubelt and Gerhard Neumann},
  title = {Learning Robust Policies for Object Manipulation with Robot Swarms},
  year = {2018},
  pages = {7688-7695},
  booktitle = {Proceedings of the IEEE International Conference on Robotics and Automation},
  crossref = {p11126},
  keywords = {swarm robotics, policy search, kernel methods, kilobots},
  url = {http://www.ausy.tu-darmstadt.de/uploads/Team/GregorGebhardt/LearningRobustPoliciesForObjectManipulationWithRobotSwarms.pdf},
}

@MASTERSTHESIS{daun_robust_slam_2017,
  author = {Kevin Daun},
  title = {Robust 3D SLAM for Mobile Search and Rescue Robots in Challenging Environments},
  year = {2017},
  school = {Technische Universitaet Darmstadt, Department of Computer Science (SIM)},
  abstract = {Rescue robot systems operate in highly challenging environments characterized by rough, possibly unknown terrain leading to non-smooth robot motions and poor odometry estimates. To enable autonomous and assisted operation, rescue robots have to localize themselves within the environment and create a map of it by employing a SLAM (Simultaneous Localization and Mapping) approach. Performing 3D laser rangefinder SLAM requires comprehensive sensor coverage of 3D space. As state- of-the-art laser rangefinders only scan a planar slice of the environment, additional motion of the sensor relative to the robot is necessary to capture a 3D scan. Thereby, separate sections of the 3D scan are captured at different times yielding distortions in the 3D scan when the robot is moving. Based on an analysis of existing approaches, this work builds on top of the Google Cartographer approach. Cartographer combines occupancy grid based scan matching with sensor observations of IMU and odometry in a joint optimization problem to resolve the scan distortion. As scan matching accumulates error, the system maintains a pose graph to perform loop closure and reduce the error of the trajectory. For 2D laser rangefinder SLAM, representing the map with Truncated Signed Distance Fields has shown to improve the accuracy of the pose estimate and the map. To allow an exchange of the map representation, a generalized formulation for generic grid maps is proposed. As specific instances of generic grid maps, occupancy grids, Truncated Signed Distance Fields and Euclidean Signed Distance Fields are ana- lyzed. Furthermore, the alignment of the robot trajectory against external references such as the Global Navigation Satellite System is examined. In the experiments comparing the grid map instances for scan matching, the signed distance field (SDF) approaches improve the accuracy of scan matching and thus the resulting trajectory and map. Furthermore, the SDF approaches yield a reduced number of iterations for the optimizations to converge leading to faster scan matching. However, the SDF approaches increase the map update time by an order of magnitude. The approach is evaluated in multiple scenarios showing the capability to handle highly challenging terrain with the corresponding incorrect odometry and high angular accelerations as shown in a scenario from the Robocup 2017 competition. The large scale behavior is validated on a ~ 500 m loop around a building in a dynamic environment. The proposed method detects the loop closure and corrects for the accumulated error.},
}

@Inproceedings{2014_rc_towards_highly_reliable,
  author = {Stefan Kohlbrecher and Florian Kunz and Dorothea Koert and Christian Rose and Paul Manns and Kevin Daun and Johannes Schubert and Alexander Stumpf and Oskar von Stryk},
  title = {Towards Highly Reliable Autonomy for Urban Search and Rescue Robots},
  year = {2015},
  volume = {8992},
  pages = {118-129},
  publisher = {Springer},
  editor = {R.A.C. Bianchi, H.L. Akin, S. Ramamoorthy, K. Sugiura},
  series = {Lecture Notes in Artificial Intelligence (LNAI)},
  booktitle = {RoboCup 2014: Robot World Cup XVIII},
  url = {http://www.springer.com/br/book/9783319186146},
  abstract = {This paper describes the approach used by Team Hector Darmstadt for participation in the 2015 RoboCup Rescue League competition. Participating in the RoboCup Rescue competition since 2009, the members of Team Hector Darmstadt focus on exploration of disaster sites using autonomous Unmanned Ground Vehicles (UGVs). The team has been established as part of a PhD program funded by the German Research Foundation at TU Darmstadt and combines expertise from Computer Science and Mechanical Engineering. We give an overview of the complete system used to solve the problem of reliably finding victims in harsh USAR environments. This includes hardware as well as software solutions and diverse topics like locomotion, SLAM, pose estimation, human robot interaction and victim detection. In 2015, the team focuses on improving the rough terrain motion capabilities of used platforms as well as manipulation capabilities. As a contribution to the RoboCup Rescue community, major parts of the used software have been released and documented as open source software for ROS.},
}

@BACHELORTHESIS{daun_collision_handling_2014,
  author = {Kevin Daun},
  title = {Collision Handling between Rigid and Deformable Bodies with Continuous Penalty Forces},
  year = {2014},
  school = {Technische Universitaet Darmstadt, Fraunhofer IGD},
}