E i n d h o v e n

Recente Publicaties

  • Olthuis, J. J. (TU/e), van der Meer, N. B. (TU/e), Kempers, S.T. (TU/e), van Hoof, C.A. (TU/e), Beumer, R.M. (TU/e), Kuijpers, W. J. P. (TU/e), Kokkelmans, A.A. (TU/e), Houtman, W. (TU/e), van Eijck, J. J. F. J. (TU/e), Kon, J.J. (TU/e), Peijnenburg, A. T. A. (TU/e), van de Molengraft, M. J. G. (TU/e) (2022). Vision-Based Machine Learning in Robot Soccer. In: Alami, R., Biswas, J., Cakmak, M., Obst, O. (eds) RoboCup 2021: Robot World Cup XXIV. RoboCup 2021. Lecture Notes in Computer Science(), vol 13132. Springer, Cham. Abstract - Robots need to perceive their environment in order to properly interact with it. In the RoboCup Soccer Middle Size League (MSL) this happens primarily through cameras mounted on the robots. Machine Learning can be used to extract relevant features from camera imagery. The real-time analysis of camera data is a challenge for both traditional and Machine Learning algorithms, since all computations in the MSL have to be performed on the robot itself.This contribution shows that it is possible to process camera imagery in real-time using Machine Learning. It does this by presenting the current state of Machine Learning in MSL and providing two examples that won the Scientific and Technical Challenges at RoboCup 2021. Both examples focus on semantic detection of objects and humans in imagery. The Scientific Challenge winner presents how YOLOv5 can be used for object detection in the MSL. The Technical Challenge winner demonstrates how to improve interaction between robots and humans in soccer using OpenPose. This contributes towards the goal of RoboCup to arrive at robots that can beat the human soccer world champion by 2050.
  • Houtman, W. (TU/e), Lopez Martinez, C.A. (TU/e & Nobleo Projects), Wang, S. (Reboocon Bionics B.V.), Ketels, A. (Speciaal Machinefabriek Ketels V.O.F.), Bruyninckx, H.P.J. (TU/e & KU Leuven), van de Molengraft, M.J.G. (TU/e) (2021). Dynamic control of steerable wheeled mobile platforms applied to an eight-wheeled RoboCup Middle Size League soccer robot. In Mechatronics, Volume 80, 2021,102693, ISSN 0957-4158, Abstract - In the RoboCup Middle Size League two teams of mobile robots play soccer against each other. During the game, agility, i.e. the ability to quickly change the direction of platform movements, is important to react or anticipate fast on the intention of opponents to efficiently perform maneuvers like ball shielding and interception. Therefore, high accelerations are desired which ideally would ask all wheels to contribute to traction in the target direction. However none of the current omnidirectional wheel-based robots in the league offers such a feature. Each pair of wheels can rotate independently about its suspension axis . The new configuration brings new challenges in control: the platform becomes kinematically nonholonomic due to the kinematic constraints around the pivot axes, but it is shown that in the context of the driving task the controller can keep the wheel configurations such that they can generate a force and torque in the directions needed by the task. Hereby, the restriction to minimize the position-error in its three degrees of freedom with respect to a predefined trajectory is relaxed by taking only the degrees of freedom relevant for the task into consideration. A cascaded control strategy is proposed that combines kinematic and dynamic control and also addresses the control-allocation problem. Compared to a full kinematic approach as typically applied on steerable wheeled systems, 2.3 times higher translational and 1.8 times higher angular velocity are demonstrated. For the translational acceleration and angular acceleration, improvement factors of 2.7 and 3.2 are achieved, respectively. The platform made a successful debut during the RoboCup Portuguese Open 2019, showing the robustness of the proposed approach.
  • Houtman, W. (TU/e), Kengen, C.M. (TU/e), van Lith, P.H.E.M. (TU/e), ten Berge, R.H.J. (TU/e), Kon, J.J. (TU/e), Meessen, K.J. (TU/e), ... van de Molengraft, R. (TU/e) (2019). Tech United Middle Size League Winner 2019. In RoboCup 2019: Robot World Cup XX (pp.517-528). (Lecture Notes in Computer Science; Volume 11531 LNAI, 2019). Springer. Tech Tech United Eindhoven Middle-Size League Winner 2019. Abstract - After the sequence of winning the RoboCup Middle-Size League (MSL) in even years only (2012, 2014, 2016, 2018), Tech United Eindhoven achieved its first RoboCup win during an odd year at RoboCup 2019. This paper presents an evaluation of the tournament and describes the most notable scientific improvements made in preparation of the tournament. These developments consist of our solution to (unforeseen) localisation problems and the improvements in the control architecture of our eight-wheeled robot. The progress in the shooting lever is elaborated as well as the advancements in the arbitrary ball-detection in order to improve our scoring during the Technical Challenge. Additionally, research towards the application of artificial intelligence in predicting the actions of opponents and recognizing the appearance of the opponent robots will be presented.
  • Van Lith, P. (TU/e), van de Molengraft, M.J.G. (TU/e), Dubbelman, G. (TU/e), Plantinga, M. (TU/e) (2019). A Minimalistic Approach to Identify and Localize Robots in RoboCup MSL Soccer Competitions in Real-time.  Minimalist MSL Robot Location 5.0 Abstract -This work provides a real-time Convolutional Neural Network to infer the team identity and location of soccer robots in the RoboCup Midsize League. It has been designed to have a fast turn-around time of less than 30 minutes between collecting robot images and an operational neural network, allowing the network to be trained in the time available between matches. This is an important feature, as the identification markers are unknown before a competition. This Fully Convolutional Network uses Global Average Pooling to generate a Class Activation Map with information about the location of every robot. A Blob detector is used to find the locations of the robots of each team which are translated into real-world coordinates, to determine the best game strategy.
    The resulting network allows for inference rates of more than 60 Hz.
  • Kon, J. (TU/e), Houtman, W. (TU/e), Kuijpers, W. (TU/e), & van de Molengraft, R. (TU/e) (2018). Pose and Velocity Estimation for Soccer Robots. Student Undergraduate Research E-Journal!, 4. Abstract - This paper details the design and real-time implementation of a planar state estimator for soccer robots. A camera system, encoders, gyroscope and accelerometer are combined in a two-stage Kalman filter through a constant acceleration model. Inflating Noise Variance is employed to handle slip and ensure convergence in stationary periods. The approach oers substantial improvement w.r.t. the old pose estimator.
  • de Koning L. (TU/e), Mendoza J.P. (CMU), Veloso M. (CMU), van de Molengraft R. (TU/e) (2018) Skills, Tactics and Plays for Distributed Multi-robot Control in Adversarial Environments. In: Akiyama H., Obst O., Sammut C., Tonidandel F. (eds) RoboCup 2017: Robot World Cup XXI. RoboCup 2017. Lecture Notes in Computer Science, vol 11175. Springer, Cham. Abstract - This work presents a pioneering collaboration between two robot soccer teams from different RoboCup leagues, the Small Size League (SSL) and the Middle Size League (MSL). In the SSL, research is focused on fast-paced and advanced team play for a centrally-controlled multirobot team. MSL, on the other hand, focuses on controlling a distributed multi-robot team. The goal of cooperation between these two leagues is to apply teamwork techniques from the SSL, which have been researched and improved for years, in the MSL. In particular, the Skills Tactics and Plays (STP) team coordination architecture, developed for centralized multi-robot team, is studied and integrated into the distributed team in order to improve the level of team play. The STP architecture enables more sophisticated team play in the MSL team by providing a framework for team strategy adaptation as a function of the state of the game. Voting-based approaches are proposed to overcome the challenge of adapting the STP architecture to a distributed system. Empirical evaluation of STP in the MSL team shows a significant improvement in offensive game play when distinguishing several offensive game states and applying appropriate offensive plays.
  • Schoenmakers, F. (TU/e), Meessen, K. (TU/e), Douven, Y. (TU/e), van de Loo, H. (TU/e), Bruijnen, D. (TU/e), Aangenent, W. (TU/e), ... van de Molengraft, R. (TU/e) (2017). Tech United Eindhoven Middle size league winner 2016. In RoboCup 2016: Robot World Cup XX (pp. 542-553). (Lecture Notes in Computer Science; Vol. 9776 LNAI). Springer. Abstract - The Tech United Eindhoven Mid-size league (MSL) team won the 2016 Championship in Leipzig. This paper describes the main progress we made in 2016 which enabled this success. Recent progress in software includes improved perception methods using combined omnivision of different robots and integrating the Kinect v2 camera onto the robots. To improve the efficiency of shots at the opponents’ goal, the obstacle detection is improved. During the tournament new defensive strategies were developed as an answer to the advanced attacking strategies that were seen during the round robins. Several statistics of matches during the tournament show the overall performance of Tech United at RoboCup 2016.
  • Kuijpers, W. (TU/e), Neves, A. J. R. (TU/e), & van de Molengraft, R. (TU/e) (2017). Cooperative sensing for 3D ball positioning in the RoboCup middle size league. In RoboCup 2016: Robot World Cup XX (pp. 268-278). (Lecture Notes in Computer Science; Vol. 9776 LNAI). Springer. Abstract - As soccer in the RoboCup Middle Size League (MSL) starts resembling human soccer more and more, the time the ball is airborne increases. Robots equipped with a single catadioptric vision system will generally not be able to accurately observe depth due to limited resolution. Most teams, therefore, resort to projecting the ball on the field. Within the MSL several methods have already been explored to determine the 3D ball position, e.g., adding a high-resolution perspective camera or adding a Kinect sensor. This paper presents a new method which combines the omnivision camera data from multiple robots through triangulation. Three main challenges have been identified in designing this method: Inaccurate projections, Communication delay and Limited amount of data. An algorithm, considering these main challenges, has been implemented and tested. Performance tests with a non-moving ball (static situation) and two robots show an accuracy of 0.13 m for airborne balls. A dynamic test shows that a ball kicked by a robot could be tracked from the moment of the kick, if enough measurements have been received from two peer robots before the ball exceeds the height of the robots.