Earthquakes pose critical challenges for search and rescue (SAR), as narrow operational time windows and hazardous terrain constrain human capabilities - highlighting the need for robotic systems. Biomimetic snake robot systems offer a promising solution with their adaptability and flexibility in navigating irregular surfaces. Prior research has largely focused on mechanical design, with limited quantitative, task-specific evaluation of gaits in post-earthquake settings. To address this gap, we compare lateral undulation and sidewinding gaits, hypothesizing that their differing biomechanics yield distinct advantages across terrains. Using the Bullet Physics engine, we model a snake robot and simulate two SAR-inspired locomotion challenges: (1) a stepped ramp with repeating incline-flat patterns simulating collapsed flooring, and (2) a randomized heightfield terrain resembling concrete rubble. Performance metrics include cost of transport, contact stability, load distribution, and either completion time (Task 1) or success rate (Task 2). Results indicate sidewinding is more energy-efficient across both tasks while lateral undulation provides markedly better contact stability and more consistent load distribution Sidewinding completed Task 1 faster and lateral undulation achieved a higher success rate in Task 2. Additional evaluations showed that directional control parameters held a critical role in improving balance and reducing slippage for both gaits. These findings support informed gait selection in real deployments and advance the development of robust SAR-oriented snake robotics.
The decontamination of legacy nuclear or other hazardous waste sites is a critical task that exposes human workers to significant risks. This paper presents a robotic system designed to mitigate human exposure by enabling remote, operator-supervised soil sampling using a retrofitted industrial walking excavator. Our integrated approach allows a single operator at a safe distance to supervise the entire soil sampling mission, from site mapping to task execution.The system’s architecture seamlessly combines (1) a cloud-based SLAM framework for generating consistent and accessible site maps; (2) a software stack for traversability estimation and navigation, enabling robust mobility in unstructured terrain; and (3) a shared autonomy manipulation framework where the operator designates a target and the robot autonomously executes the complex, safety-critical soil sampling motion.This paper outlines the system design and reports on the successful validation of its core components. In a local testbed designed to simulate hazardous waste sites, we validated several key systems on the semi-autonomous excavator: the cloud-based SLAM pipeline, local 3D reconstruction for real-time environment modeling, intuitive Cartesian control of the end-effector, the autonomous sampling workflow, and the navigation stack. Field experiments subsequently validated all necessary components for the complete sampling workflow, demonstrating the viability of our shared autonomy approach and establishing a promising foundation for future field deployment.
This paper presents a robotics framework to support emergency entities, integrating a large language model-based multi-agent system (LLM-MAS) to manage the movements of uncrewed ground vehicles (UGVs) in search and rescue (SAR) operations. The system allows operators without robotics expertise to issue natural language voice commands to autonomous UGVs, ensuring their safe and context-aware navigation based on real-time mission data. The proposed ROS 2-based communication architecture integrates UGVs into existing SAR command hierarchies, featuring robust methods for message acknowledgment, prioritization, and operational window synchronization that enable effective coordination between human teams and autonomous robots. During the XIX Workshop on Security, Emergencies, and Catastrophes (Malaga, Spain), over 150 first responders participated in six parallel SAR missions, with some teams providing their real-time positional data via equipment-integrated sensors. The LLM-MAS used this data to make autonomous decisions on the ground, such as determining which SAR team the UGV should assist, and then sent the selected target location to an external path planner through a tool call. The planner generated a route to the target, which the UGV followed while its movements were monitored in real time through a web interface. The system successfully executed material transport tasks across three critical service routes that supported the six parallel missions. Direct feedback from first responders confirmed that our system improved mission efficiency and was praised for its intuitive, responsive performance in coordinated decision-making.
Reliable communication is critical for robotic teleoperation in Search and Rescue missions, particularly in challenging environments such as street canyons, underground garages, and industrial halls. While mesh networks, operating within unlicensed frequencies, extend range through multi-hop connectivity, they face interference, frequency competition, and reduced data rates per hop. Cellular technologies offer a promising alternative by leveraging dedicated frequency bands. However, standard-compliant Integrated Access and Backhaul (IAB) solutions are limited by hardware compatibility restrictions that result in vendor lock-in. This paper introduces LANTERN, an open multi-radio relay system that builds on the established SEAMLESS Multi-Link protocol, enabling interoperability across diverse communication technologies, including cellular and mesh networks. This architecture ensures reliable operation even in complex environments by combining established systems with modern ones without requiring substantial infrastructure changes. Initial validation experiments using a homogeneous 5G over-the-top relay demonstrate seamless transitions between the umbrella and relay networks, enabling uninterrupted immersive robotic teleoperation during mission-critical operations. In addition, a second case study demonstrates the functionality of heterogeneous relaying between a cellular umbrella and a mesh network, while highlighting potential future improvements to enhance handover and link usage when dealing with heterogeneous technologies.
This paper presents a heterogeneous aerial-ground robotic system for reliable communication and teleoperation in CBRNe (Chemical, Biological, Radiological, Nuclear, and Explosive) environments. The system follows a human-in-the-loop design where an operator teleoperates a ground vehicle equipped with a robotic manipulator, while multiple UAVs maintain wireless connectivity. One UAV is tethered to the UGV through an optical fiber providing a high-bandwidth link, and others form a Wi-Fi HaLow mesh network. To the best of the authors' knowledge, Wi-Fi HaLow has not yet been experimentally validated for real-time robotic teleoperation. Experiments demonstrated stable real-time teleoperation at a 300m distance, confirming the system's reliability in communication-challenged environments.
Incidents in chemical plants can pose a high level of risk and harsh environments for first responders. Contamination and explosion hazards can deny human access to the affected infrastructure, underscoring the need for capable robot systems. This field report documents the successful deployment of a robotic task force to neutralize an explosive gas hazard at a chemical plant after a fire incident. An Unmanned Ground Vehicle (UGV) with a custom manipulation tool opened a critical valve under hazardous conditions, averting the threat of a large-scale explosion. We provide insights into robot deployment and use the mission results to highlight both the importance of rescue robotics and limitations of using research platforms in real emergency deployments, such as communication constraints and the need for enhanced operator-assistance functions.
Developing robust perception systems for autonomous Search and Rescue (SAR) robots remains a critical challenge due to the scarcity of annotated data from hazardous environments. Real-world datasets are constrained by safety limitations and rarely capture the visual complexity, structural variability, and sensor artifacts characteristic of disaster scenarios. This work introduces a reproducible methodology for generating high-fidelity RGB datasets using NVIDIA IsaacSim, enabling the photorealistic simulation of post-disaster environments with controllable terrain textures, occlusions, and dynamic sensor effects. The pipeline incorporates automatic pixel-wise semantic labeling and is used to train a brain-like neural network model called Bayesian Confidence Propagation Neural Network (BCPNN) that first learns representations in an unsupervised manner and then classifies terrain textures once the labels are made available. The proposed framework is validated through extensive experiments in both simulated environments and physical testbeds that recreate representative SAR conditions under controlled settings. Results show that BCPNN models trained exclusively on synthetic data can generalize effectively to real-world RGB inputs, capturing terrain semantics with sufficient fidelity to support autonomous mobility decisions. This contribution provides a scalable data generation and learning pipeline for perception in extreme environments and establishes a practical foundation for deploying probabilistic terrain understanding in field-ready SAR robotics. The datasets of this development are available in the Zenodo repository. and code in the GitHub repository.
Control policies that can generate complex robot behaviors can be learned from human demonstrations; this is especially useful for learning dexterous object manipulation tasks common to field and rescue robotics applications. However, training data collected from human demonstrators often exhibit significant variations in quality, with poor quality demonstrations potentially compromising the performance of the learned policy. This paper introduces a novel semi-supervised learning framework that filters high-quality trajectories from a mixed-quality dataset for policy learning. Our method addresses two challenges in autonomous robotics: making the most of limited expert demonstration data, and; the need to capture non-Markovian decision-making characteristics inherent in human expert behavior. Our proposed framework leverages task-specific pre-training combined with semi-supervised clustering for quality discrimination. A transformer architecture is used to capture temporal dependencies and task context. Experiments in simulation using the Franka Emika Panda robot arm performing generic tasks relevant to field and rescue robotics, such as lifting, transporting, and placing, show that our method enables subsequent imitation learning and reinforcement learning algorithms to learn better control policies with less data, yielding an average improvement of 4.8% in task completion success rate across the Lift, CanPickPlace, SquareNuts Robomimic benchmarks tasks. The project page is available at https://github.com/RedmondLabUCD/Filtering-Human-Demonstration-Datasets-to-Improve-Policy-Learning-for-Robotic-Manipulation.
This paper presents a fire detection and ranging pipeline for autonomous firefighting drones designed to extinguish fires in areas difficult to reach, such as at elevation or in remote forest areas. Specifically, we propose a real-time pipeline for the detection and ranging of fire, including the detection of people for safety. We compare different object detection models for the localization of fire, smoke and humans and show that fusing the Red, Green of RGB images and the thermal image in a context-aware manner results in a more robust fire detection. Finally, we present a method to localize the source of the fire in RGB images through morphological operations in the HSV color space when thermal radiometric data is not available. Our pipeline is trained, tested and validated on public and custom datasets.
Urban decontamination tasks are hazardous, complex, and time-critical, often requiring human intervention in environments with diverse geometry and unpredictable conditions. This work presents CERBERUS (Collaborative Emergency Response for Biohazard Environmental Remediation via Uncrewed Systems), a novel, field-tested multirobot platform that integrates an uncrewed ground vehicle (UGV), uncrewed aerial system (UAS), and a robotic manipulator to deliver spray-based decontamination and surface sampling across multi-elevation urban environments. Designed for modularity and robustness, CERBERUS combines high payload capacity, extended operational time, and precise surface interaction through coordinated control of its ground and aerial subsystems. A custom-built spraying and sampling system enables decontamination across complex geometries, from kerbs and door handles to rooftops and overhangs. Field trials conducted in a full-scale urban training compound demonstrated the system’s ability to navigate obstacles and challenging terrain, apply decontaminant in varied modes, and collect environmental swabs. The results highlight both the feasibility and the challenges of deploying multi-modal autonomous systems in high-consequence environments. This paper outlines the technical architecture, field performance, and future development roadmap towards greater autonomy and operational resilience for deployment in civil protection and national recovery scenarios.
Simultaneous localization and mapping is unreliable for realizing autonomous navigation and inspection in underground environments. Therefore, a tunnel-patrol robot system that can simultaneously accomplish both tasks is proposed in this paper. The robot is equipped with a 3D light detection and ranging device and an RGB camera; the acquired point clouds are converted into bird’s-eye-view images showing traversable areas. A set of candidate waypoints is supplied to a large multimodal model (LMM). Customizing the prompts enables the LMM to select an optimal waypoint and generate the corresponding velocity command, thereby achieving map-free navigation. A multi-persona deliberation prompt called "MAGI-ToT" introduces conditional branching behaviors triggered by abnormal-object detection. Field experiments conducted in both muddy and concrete-paved tunnels showed that the system continuously avoided walls and ditches while traveling and halted upon recognizing a safety helmet. These results confirm that the LMM can achieve flexible path selection alongside diverse inspection tasks.
The increasing frequency and severity of wildfires, industrial accidents, and urban blazes have underscored the urgent need for innovative firefighting technologies that enable rapid and localized intervention. This paper presents a systems-engineering-based development of the Impulse Fire Fighting System (IFFS) Drone, a compact, modular, and cost-effective aerial platform that integrates an impulse-based extinguishing mechanism within a fully self-contained system architecture. Unlike existing aerial firefighting concepts that rely on large-scale liquid delivery or tethered water sources, the IFFS Drone employs a high-impulse discharge system capable of propelling atomized extinguishing agents with minimal volume consumption, everything contained onboard.The developed platform demonstrates stable flight despite exposure to recoil forces of approximately 1 kN per shot, achieved through both unique hardware and control design. The design process follows a systems engineering approach adapted for UAV development, integrating model-based analysis, iterative prototyping, fine-tuning, and field validation.The presented work highlights the complete design process, subsystem characterization, control and structural design integration, and field deployment across representative operational scenarios, including high-rise structures, bushfire environments, and rooftop fires involving photovoltaic systems. The results demonstrate that impulse-based suppression at drone scale is both feasible and effective, paving the way for future autonomous and networked firefighting UAV systems capable of rapid intervention in high-risk environments. A complementary video showcases the key experiments and operational demonstrations of the IFFS drone.
In the decommissioning of the Fukushima Daiichi Nuclear Power Station operated by Tokyo Electric Power Company, a robot has been developed to remotely accumulate zeolite sandbags placed in the underground floors. Since the accumulation work of the sandbags would be carried out over several months, the radiation tolerance of the electrical components installed on the robot affects its lifespan. An investigation into the radiation tolerance of the electrical components revealed that the controller which is functionally important tends to fail relatively early. This is assumed to be due to the high integration of the semiconductors used in the controller. Therefore, we developed a remote-control system, placing a highly integrated Programmable Logic Controller as the controller in a low radiation area, whilst equipping the robot operating in high radiation environments with remote I/O that has lower integration. The following irradiation test had shown that the radiation tolerance of the remote I/O is higher than that of the PLC, enabling the lifespan of the robot to be extended by 1.8 times. This robot has been deployed to the actual site and implemented the zeolite accumulation work.
This paper investigates the benefits of immersive and intuitive 3D virtual reality (VR) control versus conventional 2D screen-based control for teleoperating a mobile manipulator in hazardous scenarios. We implemented two functionally similar control methods on a ruggedized response robot: (1) a mouse-driven 2D GUI with multiple live video streams and a 2D projection of environment and robot, and (2) a Godot-based walkable VR environment operated with a commercial VR headset, featuring an interactive digital twin and overlaid sensor data. In a within-subjects experiment with nine participants (with robot control experience) performing inspection and manipulation tasks, we measured objective performance (time on task, accuracy) and subjective responses with standardized questionnaires (SSQ, SEQ, ASQ, UMUX, NASA-TLX). Both control methods were well suited for the given tasks, but participants predominantly preferred the VR control. Without inducing excessive simulator sickness, VR control yielded statistically significant 12% faster task completion, 45% lower mental workload, and higher usability ratings. These results demonstrate that intuitive VR interfaces can enhance expert operators’ performance and the usability of response robots with a manipulator in time-critical scenarios.
Fuel debris retrieval is the indispensable task for decommissioning Fukushima Daiichi Nuclear Power Station (FDNPS). One of the significant barriers is the high radiation which deteriorates the operation efficiency. The radiation makes the robot sensors and the electronics get out of order easily. For the acceleration of decommissioning operation, it is required to develop the DX (Digital Transformation) platform to evaluate the sensor performance in the radiation environment. This research presents the approach exploiting a photorealistic game engine. The developed DX platform contributes to evaluate the sensor degradation and image processing algorithm inside the PCV (Primary Container Vessel) of FDNPS.
The range of applications for mobile robots is continuously growing. Affordable robots are increasingly being used in fields such as healthcare, restaurants, and as assistants in shopping malls. In indoor environments, a robot usually needs to localize itself within a previously recorded map in order to perform its tasks. To do so, it detects features in its surroundings using environmental sensors, compares them with features in the map, and thereby estimates its position within the map. This approach works well as long as a sufficient number of features are present in the robot’s environment. However, there are situations in which too few features are available for accurate localization. In open spaces that exceed the range of the sensors, the robot cannot perceive any features. Similarly, long corridors with parallel walls pose a problem for determining the position along the wall, and circular rooms can degrade the estimation of orientation. Detecting such situations, where localization becomes unreliable due to a lack of features, would be desirable, since the robot could then rely more strongly on other sensor information such as wheel odometry or the IMU. In this paper, we present a new approach that improves the localization of mobile robots when they temporarily encounter environments with only a few detectable features. The localization algorithm NDT-AMCL is extended by a method that analyzes the environment sensor data to assess the expected localization quality and adjusts the confidence in the wheel odometry accordingly.
Corrosion of steel infrastructure is an extremely costly problem for human society, with substantial financial, environmental, and safety implications worldwide. Standard mitigation efforts require manual inspection and repair, which are labor-intensive, dangerous, and complicated by limited access to facilities and affected material. To address this problem, this paper documents the systematic evolution of an integrated robotic approach for autonomous corrosion detection and treatment in complex industrial environments. Our approach combines image and LiDAR-based perception with geometry-aware planning to generate and execute surface coverage trajectories while applying a spray-based corrosion inhibitor. We also introduce a compliant virtual fixture generation method that improves reachability and reliability in these geometrically constrained environments. We demonstrate our approach in trials with a dual-arm mobile robot in a decommissioned oil and gas processing facility. Based on limitations discovered with the scalability of this initial system, we present a revised heterogeneous team that unifies a series of targeted research efforts to provide a more comprehensive solution. Field tests demonstrate the capability to detect and treat corrosion on small, irregular surfaces representative of real-world infrastructure, confirming the feasibility and potential of early-stage in situ robotic corrosion mitigation.
We propose a multi-target human tracking framework for outdoor facility monitoring using a sparse heterogeneous sensor network consisting of static RGB-D cameras and mobile LiDAR-equipped security ground robots. In contrast to existing approaches limited to short-term occlusions, our method maintains tracking consistency during extended coverage gaps by using a novel motion model. The model combines intention-aware global path planning with the Social Force Model to generate realistic trajectory hypotheses. Using those predictions, we solve the tracking problem with a particle filter that estimates the current and future distribution of human positions based on movement history and metric semantic map data. Experiments on real trajectory data from a commercial facility and synthetic data demonstrate superior performance compared to several baselines. The results show the effectiveness of intention-aware prediction using metric semantic map data for maintaining tracking consistency in sparse sensor environments.
Navigating autonomous ground robots through dense and unstructured environments, such as forests, remains a major challenge due to the complexity of natural terrains and the computational burden of traditional navigation pipelines. Conventional approaches often rely on detailed prior maps, rigid rule-based systems, and computationally heavy sensor fusion techniques, which tend to lack generalization across varying environments.In this work, we present a reinforcement learning (RL) strategy for mapless navigation of ground robots in dense forested areas, relying solely on a 3D LiDAR sensor and the position of the target and the agent. The agent is trained using Proximal Policy Optimization (PPO) within a photorealistic simulation framework tailored for outdoor navigation tasks. We adopt a curriculum learning scheme that incrementally increases obstacle density during training.Experimental results in simulation show that the trained agent is capable of navigating challenging forest scenarios effectively, consistently reaching target locations even under high obstacle density. Evaluation based on Success Rate and Success Path Length (SPL) metrics highlights the robustness and adaptability of the learned policy, underscoring its potential for real-world deployment on resource-constrained robotic platforms.
Landslides cause $2-4 billion in annual damages and 4,500 fatalities globally, yet current satellite monitoring lacks the resolution and temporal flexibility for early detection. We present an autonomous robotic platform with advanced perception integrating hyperspectral VNIR/SWIR imaging, 16-channel LiDAR, and RGB camera system for comprehensive slope stability assessment. The system enables soil moisture quantification, vegetation stress detection, and deformation monitoring through novel sensor fusion algorithms targeting both creep and earth flow landslides. The platform operates in manual, teleoperated, or autonomous modes to accommodate hazardous terrain conditions. Preliminary validation demonstrates data processing capabilities and successful terrain traversability.
