Articles and essays about various AI topics that I wrote or contributed to.
Back to index page...
Fast Convergence for Object Detection by Learning how to Combine Error Functions
Conference paper
Benjamin Schnieders, Karl Tuyls (August 2018) cite
  title={{Fast Convergence for Object Detection by Learning how to Combine Error Functions}},
  author={Schnieders, Benjamin and Tuyls, Karl},
  booktitle={2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
Abstract In this paper, we introduce an innovative method to improve the convergence speed and accuracy of object detection neural networks. Our approach, Converge-Fast-Auxnet, is based on employing multiple, dependent loss metrics and weighting them optimally using an on-line trained auxiliary network. Experiments are performed in the well-known RoboCup@Work challenge environment. A fully convolutional segmentation network is trained on detecting objects' pickup points. We empirically obtain an approximate measure for the rate of success of a robotic pickup operation based on the accuracy of the object detection network. Our experiments show that adding an optimally weighted Euclidean distance loss to a network trained on the commonly used Intersection over Union (IoU) metric reduces the convergence time by 42.48%. The estimated pickup rate is improved by 39.90%. Compared to state-of-the-art task weighting methods, the improvement is 24.5% in convergence, and 15.8% on the estimated pickup rate. Continue reading...
NOctoSLAM: Fast Octree Surface Normal Mapping and Registration
Conference paper
Joscha Fossel, Karl Tuyls, Benjamin Schnieders, Daniel Claes, Daniel Hennes (September 2017) cite
  title={{NOctoSLAM: Fast Octree Surface Normal Mapping and Registration}},
  author={J. Fossel and K. Tuyls and B. Schnieders and D. Claes and D. Hennes},
  booktitle={2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
Abstract In this paper, we introduce a SLAM front end called NOctoSLAM. The approach adopts an octree-based map representation that implicitly enables source and reference data association for point to plane ICP registration. Additionally, the data structure is used to group map points to approximate surface normals. The multi-resolution capability of octrees, achieved by aggregating information in parent nodes, enables us to compensate for spatially unbalanced sensor data typically provided by multi-line lidar sensors. The octree-based data association is only approximate, but our empirical evaluation shows that NOctoSLAM achieves the same pose estimation accuracy as a comparable, point cloud based approach. However, NOctoSLAM can perform twice as many registration iterations per time unit. In contrast to point cloud based surface normal maps, where the map update duration depends on the current map size, we achieve a constant map update duration including surface normal recalculation. Therefore, NOctoSLAM does not require elaborate and environment dependent data filters. The results of our experiments show a mean positional error of 0.029 m and 0.019 rad, with a low standard deviation of 0.005 m and 0.006 rad, outperforming the state-of-the-art by remaining accurate while running online. Continue reading...
Grey Literature, Theses:
Distributed Mapping and Path Planning
Research proposal
Benjamin Schnieders (September 2015) cite
  author = {Schnieders, Benjamin},
  title = {{D}istributed {M}apping and {P}ath {P}lanning},
  howpublished = {Retrieved from},
  year = {2015}
Abstract Mapping the environment and planning paths is a substantial part of maintaining the ability to move for mobile robots. Robots typically obtain map data through a technique called simultaneous localization and mapping (SLAM), and use path planning algorithms on the generated map to find routes to arbitrary locations. This research proposal evaluates the possibilities and advantages of extending the well-established, mostly single-agent based SLAM and path planning algorithms towards the domain of multi-agent systems.

By taking advantage of increased overall computational power and storage capabilities of a multi-agent system, larger or more detailed maps can be stored than possible on a single robot. Multiple agents can cooperate in planning a path for a single agent in order to speed up the pathfinding procedure. Spreading the map over multiple agents requires a distributed path planning procedure in order to find a path without assembling a monolithic map. Systems of multiple agents are less susceptible to single points of failure by adding additional redundancy, thus enhancing the overall robustness of the system. Incorporating inputs originating from different types of sensors to a navigable map, a diverse robotic setup can be achieved while retaining robustness and redundancy.

Practical applications of distributed slam emerge whenever a single-robot setup is prone to complete failure, i.e., where the robot is endangered by its environment and cannot be replaced easily. In these situations, which include extraplanetary rovers or surgical microbots, multiple, cooperating robots may be used instead. Systems of multiple robots, especially those designed to work as a robot swarm, such as unmanned aerial vehicles looking for survivors in a disaster aftermath, benefit from distributed SLAM naturally, as it reduces their individual memory usage for navigation, allowing more important data to be stored or lager regions to be explored. Continue reading...
aiPath: Movement Planning in Physically Constrained Domains
Master's thesis
Benjamin Schnieders (July 2014) cite
  author = {Benjamin Schnieders},
  title = {{aiPath: Movement Planning in Physically Constrained Domains}},
  school = {Maastricht University},
  year = {2014},
  month = {July},
  note = {Retrieved from}
Abstract While pathfinding is an extensively studied field in computer games, planning the actual movements to be performed while following the path is an open field to more in-depth research. Following a path that was planned without considering optimal action selection and physical possibility might produce visible movement artifacts, which in turn can be noticed by human players. Detecting irrational, impossible or simply uncanny behavior easily triggers a repulsive reaction in the player, destroying believability of the game world, and with it the created immersion. Physical movement planning is a technique to plan an optimal path while regarding physical possibility of actions at all times, annihilating movement anomalies. This thesis research examines the use of AI search techniques to plan a path in the physical state space of a vehicle. Different discretization strategies are proposed. Two search algorithms are collaborating to find a path in the discretized physical state space. An initial probing search step makes use of a greedy best-first search in order to quickly expand nodes towards the goal. Both algorithms make use of a custom heuristic evaluation function guiding them towards the goal. An integration into a game making heavy use of open street traffic, Emergency 5, provides detailed results in a real-life scenario. Continue reading...
Designing a Steering System for Vehicular Traffic for Emergency 5
Internship report
Benjamin Schnieders (January 2014) cite
  author = {Schnieders, Benjamin},
  title = {{Designing a Steering System for Vehicular Traffic for Emergency~5}},
  howpublished = {Retrieved from},
  year = {2014},
  note = {Internship Report for Maastricht University}
Abstract This internship report describes design and implementation of a steering system for disaster simulation game Emergency 5 (EM5). The intern should focus on improving the navigation capabilities, writing re-usable code that can also work for other projects of the company. The produced steering system shows to be capable of maneuvering entities in a lifelike manner along their paths. The steering system with small improvements will subsequently be in use for the upcoming title EM5. Continue reading...
Analysis of Publication Data in Nanotechnology
Benjamin Schnieders, Dries de Rydt, Emanuel Oster, Ruben Schwarzwald (June 2013) cite
  author = {Schnieders, Benjamin and Rydt, Dries de and Oster, Emanuel and Schwarzwald, Ruben},
  title = {{Analysis of Publication Data in Nanotechnology}},
  howpublished = {Retrieved from},
  year = {2013}
Abstract Nanotechnology is a field of research concerning technology that is typically made of structures below the size of 100 nanometers. Analysis of large-scale text databases of publication data in nanotechnology can produce additional knowledge and insights not directly derivable from single publications. This report provides an overview of the methods available in general and applied on a database containing about 300.000 publications, then presents results obtained by exploratory search and answers certain research questions. In the following, these results are discussed and conclusions are drawn. Continue reading...
Automatic Language Identification
Benjamin Schnieders, Florian van Daalen (June 2013) cite
  author = {Schnieders, Benjamin and Daalen, Florian van},
  title = {{Automatic Language Identification}},
  howpublished = {Retrieved from},
  year = {2013}
Abstract Automatic language identification is an important first step for natural language processing. This paper presents an approach that breaks down any text to sequences of 2 characters, so called bigrams, and analyzes their occurrences in different languages. Detection proceeds likewise, choosing the most likely language. It will show that, using statistical methods, even single bigrams can hint towards a language classification, and that by combining information of multiple bigrams, an extremely performant and accurate classifier can be built. Continue reading...
Graph-based Simultaneous Localization and Mapping on the TurtleBot Platform
Rik Claessens, Yannick Müller, Benjamin Schnieders (January 2013) cite
  author = {Claessens, Rik and Müller, Yannick and Schnieders, Benjamin},
  title = {{Graph-based Simultaneous Localization and Mapping on the TurtleBot Platform}},
  howpublished = {Retrieved from},
  year = {2013}
Abstract Simultaneous localization and mapping (SLAM) is a non-trivial task in robotics. SLAM is an algorithm that generates a map of an environment that is previously unknown to the robot while at the same tracking the location of the robot within that environment. Using the TurtleBot platform and the Robot Operating System (ROS), a Graph-based SLAM algorithm (GraphSLAM) is successfully implemented and discussed. Additionally, an exploration strategy is implemented to allow the robot to explore and map the environment autonomously. Continue reading...
Alpha-Beta Search in Pentalath
Benjamin Schnieders (December 2012) cite
  author = {Schnieders, Benjamin},
  title = {{Alpha-Beta Search in Pentalath}},
  howpublished = {Retrieved from},
  year = {2012}
Abstract This article presents general strategies and an implementation to play the board game Pentalath. Heuristics are presented, and pruning improvements to the alpha-beta framework are tested. The resulting program will be able to play Pentalath on a challenging level. Continue reading...
Find more information about Pentalath by Cameron Browne
Event based collision detection
Benjamin Schnieders (July 2012)
Abstract This web article presents an even-driven update logic for many spacial simulations, which only updates if certain events are received. It is compared to a common frame-based update system. Using a simple particle collision model, both approaches are presented and occurring errors are demonstrated. With showing which one to prefer in certain situations, the article is concluded. Read more...
Humanoid robots: will they ever be able to become like us, and if so, do we want this to happen?
Benjamin Schnieders (May 2011)
Abstract This essay shortly discusses the question whether there will be humanoid robots among us in the next few years, and the ethical consequences, if this should be the case. Read more...
Using STRIPS for automated computer game scene generation
Benjamin Schnieders (July 2008)
Abstract To be able to provide harmonious artificial game worlds, game designers have to write huge, unflexible scripts defining NPC behavior. This paper investigates to what extend it is possible to use a STRIPS planner to generate NPC plans of a scene automatically during runtime. An experiment was performed in a simulated game environment, and test runs show the ability of generating and executing plans. It was found that even from simple scene settings, many plots can be generated, in which the agents show intuitive, cooperative and emergent behavior. From the results it can be concluded that STRIPS planners are suitable for planning scenes and that the technique presented is an improvement over commonly used scripting techniques. Read more...