Reliable odometry in highly dynamic environments remains challenging when it relies on ICP-based registration: ICP assumes near-static scenes and degrades in repetitive or low-texture geometry. We introduce Dynamic-ICP, a Doppler-aware registration framework. The method (i) estimates ego translational velocity from per-point Doppler velocity via robust regression and builds a velocity filter, (ii) clusters dynamic objects and reconstructs object-wise translational velocities from ego-compensated radial measurements, (iii) predicts dynamic points with a constant-velocity model, and (iv) aligns scans using a compact objective that combines point-to-plane geometry residual with a translation-invariant, rotation-only Doppler residual. The approach requires no external sensors or sensor-vehicle calibration and operates directly on FMCW LiDAR range and Doppler velocities. We evaluate Dynamic-ICP on three real-world datasets-HeRCULES, HeLiPR, AevaScenes-focusing on highly dynamic scenes. Dynamic-ICP consistently improves rotational stability and translation accuracy over the state-of-the-art methods.

Automation of healthcare workflows and devices demands safe and trustworthy robotic behavior, particularly in environments shared with patients and medical staff. For ceiling-mounted imaging robots, the key challenge lies in perceiving and monitoring the 3D workspace to plan safe, collision-free motions around people and equipment. Beyond simple obstacle avoidance, semantic understanding is essential to distinguish between object types — such as patients, walking aids, or medical tools — and to adapt motion behavior accordingly. We address this challenge with a semantic-aware obstacle tracking and avoidance pipeline that extends prior 2D semantic navigation concepts into full 3D space. The approach combines 2D semantic segmentation with depth projection to estimate object positions and dimensions in real time from RGB-D data. These detections are fused in a tracking module to build a continuous, semantic world model from which class-dependent safety margins are derived. The resulting information enables adaptive motion planning that increases distance from high-risk objects (e.g., persons) or reduces velocity when close interaction is required. Experiments on a real ceiling-mounted robot in laboratory scenarios demonstrate the system’s ability to enhance safety, predictability, and contextual awareness during automated healthcare procedures.

Autonomous medical systems must meet stringent hygiene and safety requirements while operating reliably in dynamic clinical environments. This paper presents a collision avoidance system based on the fusion of two mm-wave radar technologies - frequency modulated continuous wave (FMCW) and pulsed coherent radar (PCR). The system is fully integrated behind sealed covers of medical devices and enables unobtrusive and hygienic use without compromising functionality. We demonstrate that both radar types provide robust detection of dynamic obstacles, even through layers of disinfectants, blood and polycarbonate materials. A fail-safe system architecture based on redundant sensor paths and dual microcontrollers ensures reliable operation under fault conditions. Experimental validation on a robotic X-ray system confirms the responsiveness, accuracy and suitability of the system for clinical integration. The results show that radar fusion offers a promising path to hygienic and certified safe motion planning in healthcare robotics.

Simultaneous localization and mapping is a critical capability for autonomous systems. Traditional SLAM approaches often rely on visual or LiDAR sensors and face significant challenges in adverse conditions such as low light or featureless environments. To overcome these limitations, we propose a novel Doppler-aided radar-inertial and LiDAR-inertial SLAM framework that leverages the complementary strengths of 4D radar, FMCW LiDAR, and inertial measurement units. Our system integrates Doppler velocity measurements and spatial data into a tightly-coupled front-end and graph optimization back-end to provide enhanced ego velocity estimation, accurate odometry, and robust mapping. We also introduce a Doppler-based scan-matching technique to improve front-end odometry in dynamic environments. In addition, our framework incorporates an innovative online extrinsic calibration mechanism, utilizing Doppler velocity and loop closure to dynamically maintain sensor alignment. Extensive evaluations on both public and proprietary datasets show that our system significantly outperforms state-of-the-art radar-SLAM and LiDAR-SLAM frameworks in terms of accuracy and robustness. To encourage further research, the code of our Doppler-SLAM and our dataset are available at: url{https://github.com/Wayne-DWA/Doppler-SLAM}.

4D imaging radars, commonly known as 4D radars, deliver comprehensive point cloud data that encapsulates range, azimuth, elevation, and Doppler velocity information even in harsh environmental conditions, such as rain, snow, smoke, and fog. However, 4D radar data also suffers from high noise and sparsity, which poses great challenges for SLAM applications. This paper presents RIV-SLAM, a complete radar-inertial-velocity optimization-based graph SLAM system designed to exploit the full potential of 4D imaging radar technology. RIV-SLAM consists of four integral components: front-end, loop closure, IMU pre-integration and graph optimization, each optimized to effectively leverage the unique attributes of radar data and tightly coupled with IMU data. This is also the first SLAM system known to us that outputs an optimized ego velocity. This capability ensures reliable ego motion estimation under extreme conditions (e.g., wheel odometry fails). Furthermore, we develop a new ground extraction approach, specifically adapted for the 4D imaging radar, which substantially improves the system’s z-axis accuracy. Comprehensive evaluations of the RIV-SLAM system on a variety of datasets demonstrate its superior performance, significantly surpassing existing state-of-the-art Radar-SLAM frameworks. The code of RIV-SLAM will be released at: RIV-SLAM

As public interest in autonomous driving systems grows, safety is becoming a critical issue. Extensive testing is therefore required before these systems can be deployed in real-world traffic. Simulation-based testing has proven to be a valuable tool for evaluating autonomous systems, but there remains a gap between simulation and real-world testing. A system may be well tested in simulation, but in real-world testing it may encounter sudden events that result in unpredictable behavior, and equipment may be damaged in the event of a malfunction. This paper describes a novel augmentation interface for Light Detection And Ranging (LiDAR) sensor data that aims to bridge this gap. The interface is designed to generate realistic test scenarios on live data streams, making it ideal for investigating special and borderline cases. The interface has been developed for the Robot Operating System (ROS) using the Point Cloud Library and is intended to be used for testing an autonomous shunting locomotive. With this interface, a data stream from a 3D LiDAR can be augmented with any given object represented by a point cloud, allowing for the use of data from both simulation and real-world environments.

Obstacle detection is crucial for ensuring the safety of autonomous robots and their surroundings in unstructured outdoor environments. Objects with minimal lateral dimensions can pose risks to the robot or serve as important elements in the infrastructure it operates in. Detecting these structures becomes particularly challenging when tall vegetation is present. Distinguishing between soft, traversable objects, such as tufts of grass, and potentially lethal solid obstacles is paramount to a robot's ability to operate. This paper presents a novel approach that focuses on point cloud generation and vegetation identification to facilitate the safe navigation of autonomous outdoor robots. Our approach uses a single multispectral stereo camera system that employs a novel stereo matching strategy based on binary descriptors for spectrally non-identical image pairs.

This paper proposes a novel approach for indoor robot localization that leverages a fusion of information from single-chip infrared (Time-of-Flight) and radar sensors. The aim of our research is the development of a cost-effective and lightweight system that can achieve high-precision robot localization. Unlike traditional localization methods based on LiDARs or cameras, our proposed system uses single-chip infrared and radar sensors to overcome the limitations of high cost and bulky hardware. Specifically, we employ a Doppler radar-based velocity motion model for the estimation of the robot's ego-motion, eliminating the need for additional sensors such as IMU or wheel encoders. Next, we describe a hybrid sensor model for single-chip infrared and radar sensors that provides robust and accurate environmental perception with dynamic outlier removal. Finally, we integrate these components into a Monte Carlo localization framework to generate accurate real-time estimation of the robot's position and orientation. This is the first time a single-chip infrared and radar fusion-based framework has been applied to robot localization, to the best of our knowledge. Through a comprehensive experimental evaluation, we demonstrate the system's high accuracy and efficiency, achieving an average localization error of 9 cm in diverse indoor environments. This remarkable performance, combined with the low-cost and lightweight nature of our proposed solution, positions it as a highly promising alternative for a wide range of applications, including robotics, smart homes, and autonomous vehicles. The significant advancements of this novel approach offer vast potential to revolutionize the field of localization, enabling more precise and cost-effective navigation systems.

This publication describes a novel approach to generic robot navigation using elevation maps based on point-region-Quadtrees. The described approach plans optimized trajectories in dependency of the robot’s morphology by detecting and rating obstacles. This is achieved by tailoring the tree based elevation map to the robot’s design. The approach and the related work, it is based on, is described in detail, experiments are provided, which verify the results.

In this paper, a novel approach is introduced which utilizes a Rapidly-exploring Random Graph to improve sampling-based autonomous exploration of unknown environments with unmanned ground vehicles compared to the current state of the art. Its intended usage is in rescue scenarios in large indoor and underground environments with limited teleoperation ability. Local and global sampling are used to improve the exploration efficiency for large environments. Nodes are selected as the next exploration goal based on a gain-cost ratio derived from the assumed 3D map coverage at the particular node and the distance to it. The proposed approach features a continuously-built graph with a decoupled calculation of node gains using a computationally efficient ray tracing method. The Next-Best View is evaluated while the robot is pursuing a goal, which eliminates the need to wait for gain calculation after reaching the previous goal and significantly speeds up the exploration. Furthermore, a grid map is used to determine the traversability between the nodes in the graph while also providing a global plan for navigating towards selected goals. Simulations compare the proposed approach to state-of-the-art exploration algorithms and demonstrate its superior performance.

This publication describes an application of a Truncated Signed Distance Mapping approach for disaster intervention in underground mine shafts through geometrical change detection of the shaft walls. The paper describes two main problems of such an approach (aligning two potentially huge point clouds and automatic change detection by comparing the reconstructed volumes) and explains in detail the proposed solution.

In this paper, a novel state machine for mobile robots is described that enables a direct use for exploration and inspection tasks. It offers a graphical user interface (GUI) to supervise the process and to issue commands if necessary. The state machine was developed for the open-source framework Robot Operating System (ROS) and can interface arbitrary algorithms for navigation and exploration. Interfaces to the commonly used ROS navigation stack and the explore_lite package are already included and can be utilized. In addition, routines for mapping and inspection can be added freely to adapt to the area of application. The state machine features a teleoperation mode to which it changes as soon as a respective command was issued. It also implements a software emergency stop and multiplexes all movement commands to the motor controller. To show the state machine's capabilities several simulations and real-world experiments are described in which it was used.

In order to allow robust obstacle detection for autonomous freight traffic using freight trains or shunting locomotives, several different sensors are required. Humans and other objects must be detected so that the vehicle can stop in time. Laser scanners deliver distance information and are popular in robotics and automation. Cameras deliver further pieces of information on the environment and are especially useful for the classification of objects, but do not deliver distance measurements. Thermal cameras are ideal for the detection of humans based on their body temperature if the surrounding temperature is not too similar. It is only the combination of these different sensors which delivers enough robustness. Therefore a sensor fusion and an extrinsic calibration has to take place. This article presents an approach fusing a 2D and an 8-layer 3D laser scanner with a thermal and a Red-Green-Blue (RGB) camera, using a triangular calibration target taking all six degrees of freedom into account. The calibration was tested and the results validated during reference measurements and autonomous and manually controlled field tests. This sensor fusion approach was used for the obstacle detection of an autonomous shunting locomotive.

This paper describes the estimation of the body weight of a person in front of an RGB-D camera. A survey of different methods for body weight estimation based on depth sensors is given. First, an estimation of people standing in front of a camera is presented. Second, an approach based on a stream of depth images is used to obtain the body weight of a person walking towards a sensor. The algorithm first extracts features from a point cloud and forwards them to an artificial neural network (ANN) to obtain an estimation of body weight. Besides the algorithm for the estimation, this paper further presents an open-access dataset based on measurements from a trauma room in a hospital as well as data from visitors of a public event. In total, the dataset contains 439 measurements. The article illustrates the efficiency of the approach with experiments with persons lying down in a hospital, standing persons, and walking persons. Applicable scenarios for the presented algorithm are body weight-related dosing of emergency patients.

A favoured sensor for mapping is a 3D laser scanner since it allows a wide scanning range, precise measurements, and is usable indoor and outdoor. Hence, a mapping module delivers detailed and high resolution maps which makes it possible to navigate safely. Difficulties result from transparent and specular reflective objects which cause erroneous and dubious measurements. At such objects, based on the incident angle, measurements result from the object surface, an object behind the transparent surface, or an object mirrored with respect to the reflective surface. This paper describes an enhanced Pre-Filter-Module to distinguish between these cases. Two experiments demonstrate the usability and show that for single scans the identification of mentioned objects in 3D is possible. The first experiment was made in an empty room with a mirror. The second experiment was made in a stairway which contains a glass door. Further, results show that a discrimination between a specular reflective and a transparent object is possible. Especially for transparent objects the detected size is restricted to the incident angle. That is why future work concentrates on implementing a post-filter module. Gained experience shows that collecting the data of multiple scans and postprocess them as soon as the object was bypassed will improve the map.

This paper proposes a novel, probability-based 2D scan matching approach called Random Normal Matching. It improves the robustness of ohm-tsd-slam - an existing 2D simultaneous localization and mapping algorithm. Using the SLAM's truncated signed distance map representation, a probability field is generated. The probability field is used to determine a pre-transformation for Iterative Closest Point scan matching. This combination results in an accurate and robust 2D SLAM, making it highly suitable for the application in rescue robotics with rough terrain. Test results show that even without processing odometry data, the proposed approach is well competitive with other state-of-the-art SLAM algorithms.

The aim of this paper is to present a decentralized control architecture for an autonomous transportation system in the manufacturing facility of the future. Each component in the factory (machine, robot, operator ... ) is represented as an individual agent on the cloud and makes autonomous decisions based on the information exchanged with other agents. Production machines control their own material replenishment by contracting the services of Autonomous Transportation Vehicles (ATV) over the cloud. Moreover, two control panels for human synergy on strategical and operational layers for monitoring and interacting with the system have been provided. Based on new trends on the industry 4.0 revolution, this paper develops an intelligent communication architecture between machines, robots and humans. This novel architecture makes the system robust, flexible and scalable. The requirements for such an architecture have been first defined and then validated via a series of theoretical cases and two experiments where communication and hardware failures have been triggered.

In order to increase the robustness of localisation and victim detection in low visibility situations it is necessary to fuse several sensors. The most common sensor used in robotics is the 2D laser scanner which delivers distance measurements. In combination with a camera the gained information can be supported by visual information about the environment. Thermal cameras are ideal for finding objects with a certain temperature, but they do not deliver distance information. The difficulty in fusing these two sensors is, that a correspondence between each distance measurement and its corresponding pixel within the thermal image needs to be found. As the laser scanner only displays one plane, this is not an intuitive task. A special triangular calibration target, covering all six degrees of freedom and being visible for both sensors, was developed. In the end the transformation between each laser scan point and its corresponding thermal image pixel is given. This allows for assigning every laser measurement within the field of view a corresponding thermal pixel. The final application will enable detection of human beings and display the distance required to reach them.

The aim of this paper is to present a low-cost Autonomous Transport Vehicle (A TV) for the transportation of material in an existing manufacturing facility. To avoid high costs of layouts and structural modifications, the ATV has to function with minimal changes to the factory. An important part of the transportation process is the docking maneuver: the action of picking up material containers from the storage areas known as supermarkets. This paper proposes three different solutions based on low-cost sensors for this docking maneuver task. First, a line-following method is introduced. To avoid the utilization of markers, the remaining solutions depend exclusively on objects available in the supermarkets. For the second approach, the common blue boxes for carrying materials are used. Third, the rails where material boxes are collected on rollers are utilized for a novel docking guidance approach. Accuracy tests of the three methods proposed expose the possibilities and limitations concerning the range of application. The box detection approach is unsuitable for current supermarkets because of its inaccuracy over larger distances. The line-following approach provides good accuracy, but requires changes to the existing supermarkets. Finally, if global localization provides the correct initial pose, the rail detection has demonstrated an accuracy similar to the line detection approach without the need of additional markers.