To overcome the severe perceptual sparsity of pipeline interiors, this paper presents an active Visual-Inertial Odometry (VIO) framework that generates its own visual landmarks on the fly. By utilizing a pulse of compressed air through a modified airbrush, the robotic platform deposits non-uniform, non-permanent fluorescent markings onto the pipe walls, creating reliable visual features for a dual-camera front-end. Alongside this active perception strategy, the system employs a multi-IMU sensor suite optimized specifically for the cylindrical manifold of pipe environments. We demonstrate via Fisher Information analysis that a closed-form, optimally placed IMU configuration significantly enhances the observability of the robot’s motion, proving that geometric arrangement dominates over mere sensor count. Validated across diverse pipe geometries and surface textures in both simulation and real-world experiments, the proposed VIO approach consistently eliminates tracking failures, achieves lower trajectory RMSE than visual-only and sub-optimal inertial baselines, and keeps the residual errors strictly contained within the pipe corridor.

This work presents a 2D human model and its use in a novel gait analysis framework, which only requires a stereo camera to produce impressive results for the inverse kinematics, inverse dynamics, as well as ground reaction forces during gait. The model is designed to resemble the human body in the sagittal plane, with anatomical landmarks used as keypoints in the inverse kinematics calculations that yield accurate estimates of the joints' motion during gait. The gait dynamics are formulated in compact form, allowing the simultaneous estimation of internal joint torques, as well as ground reaction forces via the solution of a fully-defined system of linear algebraic equations. The proposed framework offers an affordable alternative to costly gait analysis systems, and can have various applications in robotics and in biomechanics.

The fresh food industry significantly depends on manual labor, which can make up to 40

Efficient and thorough vineyard inspection is crucial for optimizing yield and preventing disease from spreading. Manual approaches are labor-intensive and prone to human error, motivating the development of automated solutions. Precision viticulture benefits greatly from access to photo-realistic 3D vineyard maps and from capturing intricate visual details necessary for accurate canopy and grape health assessment. Generating such maps efficiently proves challenging, particularly when employing cost-effective equipment. This paper presents a novel vineyard inspection and 3D reconstruction framework implemented on a Robotic Platform (RP) equipped with three stereo cameras. The framework's performance was evaluated on an experimental synthetic vineyard developed at NTUA. This testing setup allowed experimentation under diverse lighting conditions, ensuring the system's robustness under realistic scenarios. Unlike existing solutions, which often focus on specific aspects of the inspection, our framework offers a top-down approach, encompassing autonomous navigation, high-fidelity 3D reconstruction, and canopy growth assessment. The developed software is available at the Control Systems Laboratory's (CSL) bitbucket repository [1].

Gait analysis is essential in many scientific fields; to study it marker-based or markerless motion capture (MoCap) techniques are used. The latter have significantly benefited from the recent rise of research in deep learning (DL) and its applications on human mesh generation. However, insufficient and suboptimal camera viewpoint selections often lead to low-grade human mesh geometries. This paper presents an approach to consistently obtain accurate human meshes using DL-based avatar reconstruction algorithms (ARAs). Our framework provides a systematic approach, utilizing a simulated environment to inform decisions on the number of cameras and their spatial configuration to achieve optimal reconstruction results. These results are enhanced through mesh evaluation, mesh alignment, and surface reconstruction to remove poorly formed geometries and artifacts. Additionally, we present a gait analysis tool, tested in simulation and reality (Fig. 1), that detects gait phase changes, extracts the significant human body joint angles and recreates the animation of the gait cycle in 3D space. The proposed approach is open-source, adjustable, and applicable to various research contexts where gait analysis is essential.

Automating the process of collecting samples (e.g., images) from a vineyard can help to monitor the condition of the grapes with precision and prevent the spreading of diseases. A critical part of this task is the development of a robust localization algorithm so that (a) a robot is able to carry out the inspection process and (b) the vine-grower knows exactly which part of the vineyard has been inspected. In this paper, we propose a novel approach for enhancing the robustness of vSLAM by utilizing multiple stereo cameras and a novel method for detecting loops in homogeneous environments based on AprilTags, where state-of-the-art approaches may find it difficult to detect them. We test the accuracy of our method using a wheeled Robotic Platform (RP) in simulation and in a synthetic vineyard developed at CSL, NTUA [1]. The developed method achieves high accuracy in the localization of the RP in the vineyard and robustness even when a featureless object covers a large part of the Field of View of one camera. The developed software is available for testing at the CSL's bitbucket repository [2].

The recent years' progress in deep learning (DL) technology has resulted in convolutional neural networks (CNNs) capable of producing fast and accurate results, with minimal data preprocessing. Currently, gait analysis (GA) is attracting the attention of the field of deep learning due to its seamless integration applicability. Our approach focuses on CNN-based GA application on healthcare and orthopaedics. Using CNNs and visual fiducial systems for recognition and 3D mesh re-construction of the human form and the floor respectively, we can virtually recreate the human-floor interaction, which can be particularly useful in the study of gait dynamics, through the per-frame gait phase classification. However, the current state-of-the-art (SOTA) is that most CNN mesh reconstruction software produces a mesh from a monocular input. The use of photogrammetry could alternatively be implemented, but would require multiple cameras and expensive equipment. Our approach aims to create a refined mesh obtained from trinocular footage, along with an interactive and easy-to-use interface.

Quadrupedal locomotion skills are challenging to develop. In recent years, Deep Reinforcement Learning (DRL) promises to automate the development of locomotion controllers and map sensory observations to low-level actions. However, legged locomotion still is a challenging task for DRL algorithms, especially when energy efficiency is taken into consideration. In this paper, we propose a DRL scheme for efficient trotting applied on Laelaps II quadruped in MuJoCo. First, an accurate model of the robot is created by revealing the necessary parameters to be imported in the simulation, while special focus is given to the quadruped’s drivetrain. Concerning, the reward function and the action space, we investigate the best way to integrate in the reward, the terms necessary to minimize the Cost of Transport (CoT) while maintaining a trotting locomotion pattern. Last, we present how our solution increased the energy efficiency for a simple task of trotting on level terrain similar to the treadmill-robot environment at the Control Systems Lab [1] of NTUA.

ATHANASIOS S. MASTROGEORGIOU*1, YEHIA S. ELBAHRAWY*2, KONSTANTINOS MACHAIRAS1, ANDRÉS KECSKEMETHY2 and EVANGELOS G. PAPADOPOULOS1 1School of Mechanical Engineering, National Technical University of Athens, 9 Heroon Polytechniou Str. 15780, Athens, Greece. E-mail: {amast, kmach, egpapado}@central.ntua.gr, http://www.mech.ntua.gr 2Faculty of Engineering, University of Duisburg-Essen, Duisburg, 47057, Germany, E-mail: {yehia.el-bahrawy, andres.kecskemethy}@stud.uni-due.de

Quadrupedal locomotion skills are challenging to develop. In recent years, deep Reinforcement Learning promises to automate the development of locomotion controllers and map sensory observations to low-level actions. Moreover, the full robot dynamics model can be exploited, but no model-based simplifications are to be made. In this work, a method for developing controllers for the Laelaps II robot is presented and applied to motions on slopes up to 15°. Combining deep reinforcement learning with trajectory planning at the toe level, reduces complexity and training time. The proposed control scheme is extensively tested in a Gazebo environment similar to the treadmill-robot environment at the Control Systems Lab of NTUA. The learned policies produced promising results.

In this paper, an outline of NTUA’s work in the framework of project INTELLICONT is presented. We describe the current state of the air-cargo handling procedures and how the autonomous system that is under development is going to simplify these and increase the overall efficiency. Important issues and challenges regarding the system's development are discussed and a preliminary design of the novel robotic platform is given. The main tasks of this platform include the autonomous motion and locking of containers with mass exceeding one tone, avoiding at the same time obstacles and surpassing terrain discontinuities. Information regarding the selected actuators and other key electrical components, such as motor drivers and sensors are provided also. The architecture of the embedded system and the specifications of the selected Central Control Unit are described, as well as the integration of the motor drivers, sensors and other peripherals with the Robot Operating System (ROS). Further details on the development of a high accuracy localization system, which is mandatory to lock the container safely to the corresponding positions are provided also. In addition, we give details regarding the locking mechanism with integrated monitoring functionalities, an important part of the system. Simulation experiments validate the selected position controller and key system specifications are highlighted based on results. Finally, recent prototype experiments conducted to verify the localization system are presented.

The systematic study of human gait dynamics has allowed medical professionals to offer personalized treatment to individuals suffering from varying degrees of gait degeneration. Currently, marker based motion capture is regarded as the gold standard of motion analysis [1][2][3]. Nevertheless, it is a very time consuming and fatiguing process, as a multitude of markers need to be carefully positioned on an individuals body. The use of Inertial Measurement Units (IMUs) has facilitated a faster process of recording limb accelerations and velocities during locomotion, allowing the reconstruction and personalized study of the gait dynamics. However, IMUs cannot be used to reconstruct the limbs’ positions, unless the individual’s initial body pose is known. Currently, the initial body pose is gained via a motion capturing systems, overturning the time benefit of the IMUs.

This paper presents simulation results obtained with a 3D model of the NTUA quadruped robot in the Webots simulation environment during slope climbing. Initially, the robot is controlled to perform pronking on level ground in order to validate the simulation environment. Gradually, the inclination is increased and simulations are conducted to discover the maximum value of positive slope the quadruped can cope with. Finally, disturbances are introduced and it is shown that the robot’s forward deceleration mainly depends on the front leg touchdown angles.

Quadruped robot locomotion is a difficult task due to the increased system complexity and its rough environment. In addition, critical stability issues emerge when considering multibody systems such as quadruped robots. Legged robots have complex dynamics and many degrees of freedom that must be well orchestrated for achieving a robust and dynamically stable locomotion pattern. Handling positive or negative slopes enhances the locomotion qualities of legged robots, but demands more from its actuation system. However, higher torque requirements have an adverse impact on a robot’s total mass. Legged robots have an advantage in dealing with various terrain types, or in handling terrain discontinuities with the use of accurate foot placement. Such systems have hybrid dynamics that are described by different sets of differential equations, according to the phase at which the robot is in (flight phase, double stance phase, etc.). Up to now, enhanced controllers, e.g. by means of computer vision [1], have been implemented for trotting on rough terrain. However, legged robots are difficult to control and as a result, they are subject to dangerous tipover instabilities. Tipover prevention criteria have been introduced aiming at prevention of dangerous situations for mobile manipulator systems [2]. Such criteria take into account tipover or rollover when operating over uneven terrain, and/or when exerting large forces or moments [6]. Figure 1: The 3D model of the NTUA quadruped in Webots 8 Simulation Environment during dynamic running. The projection of the body’s CoM is within the robot’s support polygon. Pronking on a 20° slope has been achieved. In our current work, we investigate stability issues of quadruped robots on positive and negative slopes subject to various disturbances. For this purpose, a 3D quadruped robot model has been implemented in Webots 8, [4]. The model is equipped with the necessary sensors (gyros, accelerometers, force sensors, laser range finder etc.) for state estimation and accurate phase triggering. The control algorithm uses sensor measurements to calculate the necessary torque and touchdown leg angles for stable pronking. Initially, the robot is controlled to perform pronking on level ground. Gradually, the inclination is increased. The quadruped’s performance is validated to be similar to [5]. We analyze how stable dynamic running can be performed as terrain morphology changes, how the quadruped’s gaits can be rearranged in order to carry out these tasks, and what makes a gait more persistent to disturbances compared to alternative ones. The answer to these key questions will help us achieve stable motions similar to mobile manipulators on various terrain types and enhance already presented tipover criteria [2]. In this work, we seek to enhance the controller in various ways by answering the previously stated questions. To this end, we examine the different support situations for quadrupeds. Firstly, during the double stance phase, the legs are in contact with the ground and form a support polygon. As a result, the force angle stability measure and time to tipover can be calculated [6]. Secondly, we study quasi-static situations, during which the projection of the robot center of mass (CoM) lies at the edge of the support polygon formed by the three legs that are in contact with the ground. Then, if the robot tends to fall toward the only leg that is not in contact with the ground, the stability can be ensured again. Finally, experiments with the NTUA quadruped show that, during dynamic running, tipover may occur when the robot rotates around the (front) left toe – right toe axis (or back left and right toes respectively). In this case, we seek solutions in which the total force acting on the CoM is pointing towards the side of the robot with a leg about to contact the ground. Overall, no single approach for every terrain or inclination exists, but instead stable running also depends on the friction between the tow and the ground as well as the compliance of the ground. Simulations will be performed for level ground and for a maximum slope of 20°. Initial experiments have been conducted with the quadruped model in Webots 8, performing pronking or bounding using a controller previously developed [3]. With this controller, forward velocity on lift-off and apex height, are maintained within desired limits. During these simulations, and using the forceangle stability measure, the time to tipover will be calculated. At ECOMASS 2015, we will present our latest results and conclusions towards an enhanced controller for stable slope climbing. In addition, we will show that dynamic running cannot be performed if ground inclination or robot’s CoM increases beyond a specific level, depending on friction, weight and torque capabilities. References [1] S. Bazeille, V. Barasuol, M. Focchi, I. Havoutis, M. Frigerio, J. Buchli, C. Semini, and D. G. Caldwell, “Vision enhanced reactive locomotion control for trotting on rough terrain,” in Technologies for Practical Robot Applications (TePRA), 2013 IEEE International Conference on, 2013, pp. 1–6. [2] E. G. Papadopoulos and D. A. Rey, “A new measure of tipover stability margin for mobile manipulators,” in Proceedings IEEE International Conference on Robotics and Automation, 1996, vol. 4, pp. 3111–3116 vol.4. [3] N. Cherouvim and E. Papadopoulos, “Use of a novel multipart controller for the parametric study of a trotting quadruped robot,” in Proceedings IEEE International Conference on Robotics and Automation, 2008, pp. 805–810. [4] O. Michel, “Webots: Professional Mobile Robot Simulation,” Int. J. Advanced Robotic Systems, vol. 1, no. 1, pp. 39–42, 2004. [5] Kontolatis I. and E. Papadopoulos, “Dynamic Running Quadruped for Crater Exploration,” in 11th International Symposium on Artificial Intelligence, Robotics and Automation in Space, (iSAIRAS), Turin, 2012. [6] D. A. Rey and E. G. Papadoupoulos, “Online automatic tipover prevention for mobile manipulators,” in Proceedings of the 1997 IEEE/RSJ International Conference on Intelligent Robots and Systems, (IROS ’97), 1997, vol. 3, pp. 1273–1278.