Uncrewed Mobile Systems for Nuclear Security

• Uncrewed mobile systems in nuclear security are diverse robotic platforms (air, land, sea, underwater) that operate autonomously to enhance safety and extend mission reach.
• They integrate advanced sensing, mapping, and radiological detection with variable autonomy, supporting surveillance, C2 augmentation, and interdiction missions.
• Field validations demonstrate high reliability, precise localization, and effective risk mitigation, underscoring their critical role in nuclear safeguards and emergency response.

Uncrewed mobile systems in nuclear security encompass a technologically diverse set of robotic platforms—spanning aerial, ground, surface, and underwater vehicles—capable of extended operation without onboard humans. These systems, deployed in functions ranging from routine facility surveys to command-and-control (C2) augmentation and strategic delivery, are central to improving operator safety, expanding operational reach, and enabling new workflows in nuclear safeguards, incident response, and deterrence stability (Horowitz et al., 2019, Pryor et al., 2024, Schwaiger et al., 2024, Sinclair et al., 31 Jan 2026, Chiou et al., 2022, Mascarich et al., 2017).

1. Definitions, Taxonomy, and Core Architectures

Uncrewed mobile systems in the nuclear domain are robotic platforms—aircraft, surface vessels, submarines, ground vehicles—that can operate for extended periods without an onboard human. They are classified along three primary dimensions: autonomy level, domain of mobility, and mission type.

• By autonomy level:
  • Supervised autonomy ("human on the loop"): routine tasks autonomously, critical actions require human approval.
  • Full autonomy ("human out of the loop"): system senses, plans, and acts without real-time human intervention in a mission segment.
  • Hybrid/optionally manned: platforms capable of both crewed and uncrewed operation, e.g., B-21 Raider (Horowitz et al., 2019).
• By mobility domain:
  • Air: UAVs (endurance 40–80 hrs; non-stop records up to 64 days), optionally-manned bombers, pseudo-satellites for nuclear command and control (NC2) relays.
  • Sea: autonomous underwater vehicles (AUVs), unmanned surface vessels (USVs), automated seabed launch platforms (e.g., "Status-6" torpedo).
  • Land: ground robots for reconnaissance, convoy escort, mobile missile launchers with automated targeting.
  • Heterogeneous multi-robot teams: floor surveyors, micro-UGVs, quadrupedal manipulation platforms (Pryor et al., 2024).
• By mission type:
  • Delivery: nuclear payload transport (future uninhabited submarine-launched vehicles).
  • Surveillance/ISR: persistent observation (multimodal UAV/UUV/USV networks).
  • C2 augmentation: early warning, emergency message dissemination, strategic automated decision support (Horowitz et al., 2019).

System architectures support varying levels of onboard and remote computation (e.g., ARM SoCs for low-power platforms, industrial PCs for heavier UGVs), operate under GNSS denial using multi-modal SLAM, and can be modular via ROS-based middleware (Schwaiger et al., 2024, Pryor et al., 2024, Thakur et al., 2019).

2. Sensing, Mapping, and Radiation Detection Modalities

Sensor payloads enable comprehensive environmental and radiological data acquisition central to nuclear security operations.

• Radiation survey payloads:
  • Scintillator-based gamma/neutron detectors (NaI(Tl), CsI(Tl), CZT, LaBr₃), air-proportional counters for alpha, thin-GM for beta (Sinclair et al., 31 Jan 2026, Pryor et al., 2024).
  • Directional and imaging detectors: Compton telescopes (angular resolution ∼10° at 662 keV), self-shielding directional spectrometers (e.g., ARDUO) for gross count and direction vector estimation (Sinclair et al., 31 Jan 2026).
  • Gamma cameras, Raman spectrometers for on-site chemical/radiological substance identification (Schwaiger et al., 2024).
• Mapping and SLAM:
  • LIDAR (solid-state, time-of-flight, multi-beam), RGB-D cameras, IMUs fused in graph-based or filter-based SLAM frameworks (e.g., LIO-SAM, RTAB-Map, EKF/UKF-based visual-inertial odometry) (Thakur et al., 2019, Schwaiger et al., 2024, Pryor et al., 2024).
  • Occupancy grid or voxel-based 3D maps; semantic layers incorporating object ID, orientation, and radiological intensity (Wang et al., 2019, Dayani et al., 2021).
  • Bayesian and GPR-based radiation source localization:

  $$\mu(x_*) = k(x_*, X)[K(X,X) + \sigma_n^2 I]^{-1}C,\quad \Sigma(x_*) = k(x_*,x_*) - k(x_*,X)[K(X,X) + \sigma_n^2 I]^{-1}k(X,X_*)$$

  with $k(\cdot,\cdot)$ an RBF kernel (Schwaiger et al., 2024).

• Calibration:

  • Intrinsic detector response (count rate vs. distance, energy calibration), extrinsic sensor registration to align radiation readings with environmental maps (Mascarich et al., 2017).

3. Control, Autonomy, and Human–Machine Teaming

Control strategies span complete teleoperation, variable autonomy, and full autonomy, with workflows determined by mission criticality, operator cognitive load, and environmental uncertainty.

• Variable autonomy:
  • Level $\alpha \in [0,1]$ is dynamically adjusted based on operator workload $W$ and situational awareness $SA$:

  $$\dot{\alpha} = k_w \cdot (W - W^*) - k_{sa}\cdot(SA^* - SA)$$

  with semi-autonomous modes engaged as operator load increases (Chiou et al., 2022).

• Supervised and fully autonomous modes:

  • Human-in-the-loop remains standard for critical nuclear tasks; "human on the loop" for routine automated operations (Horowitz et al., 2019).
  • Local/global navigation (e.g., Dijkstra/A*, DWA), teach-and-repeat fallback, exploration gain-maximizing planners, receding horizon mapping (Schwaiger et al., 2024, Vannini et al., 2 May 2025, Mascarich et al., 2017).
• Human–machine interfaces:
  • GUI(s) presenting map overlays, radiation contour sliders, sensor data fusion, live 3D and video streams (Schwaiger et al., 2024, Dayani et al., 2021).
  • VR interfaces enabling semi-autonomous multi-waypoint mission planning and immersive data visualization for single-operator control (Dayani et al., 2021).
• Heterogeneous and fleet coordination:
  • Auction-based task assignment, multi-robot map fusion requiring manual or automated relocalization, dynamic role arbitration via frameworks like TeMoto (Vannini et al., 2 May 2025, Pryor et al., 2024).

4. Operational Workflows, Performance, and Field Validation

Field exercises and laboratory trials demonstrate the operational performance, coverage, and system-level reliability of uncrewed mobile systems.

Platform/Scenario	Coverage Area (m²)	Loc. RMSE (m)	Radiation Loc. Error (m)	Mission Time (min)	Key Outcomes
UGV-CBRN (Scenario 1)	180	0.12 ± 0.05	0.3 ± 0.1	14.2 ± 1.3	100% sample/valve success (Schwaiger et al., 2024)
RRSS (µUGV floor sweep)	100	~0.05 (SLAM)	<10% α detection error	30	95% floor coverage, no mission aborts (Pryor et al., 2024)
sUAS 3D radiation map	∼10×10 grid	—	0.12 (mean)	12–19	8–9% intensity error, single-op (Dayani et al., 2021)

• Performance metrics include localization RMSE, coverage ratio, dose estimation, sample retrieval and valve actuation rates, mapping resolution, and radiation localization error (Schwaiger et al., 2024, Pryor et al., 2024, Dayani et al., 2021).
• Operator load: cognitive demand is high for extended teleoperation; variable autonomy and intuitive interfaces alleviate errors and fatigue (Chiou et al., 2022).
• Experimental results: All three platforms (UGV-CBRN, RRSS, VR-enabled sUAS) achieved high rates of task success (sample, valve, or survey), sub-meter localization, and near-complete coverage on representative field challenges.

5. Risk, Reliability, and Nuclear Strategic Stability

Robustness, accident risk, and escalation pathways are central in nuclear-domain deployments.

• Reliability models:
  • System reliability $R_a(t) = \exp(-\lambda_a t)$, with human-on-the-loop provision reducing the effective failure rate: $\lambda_e = \lambda_a \cdot (1-\mu)$ (Horowitz et al., 2019).
  • Probability of accidental/unauthorized launch

  $P_a(\alpha) = P_0 + k \cdot \alpha^n$

  where $P_0$ is baseline human error, $k$ and $n$ encode additional automation risk. Example: human-only baseline accident rate in nuclear C2 ≈ $10^{-6}$/yr; imperfect automation can double this (Horowitz et al., 2019).

• Automation bias and escalation:

  • Documented operator trust failures (e.g., Oko/Petrov, Patriot fratricide) and feedback loop pathologies (VRYAN incident) demonstrate that transparency, manual-reversion capability, and automation-failure training are pivotal.
  • Full autonomy in kinetic/nuclear action remains limited; positive human authorization is enforced in all current C2 architectures.
• Operational risk mitigation:
  • Technical: redundant communication, human-readable interfaces, rigorous adversarial testing, multi-channel fail-safe mechanisms (Horowitz et al., 2019, Chiou et al., 2022).
  • Policy: arms control on uncrewed nuclear-armed platforms, transparency and certification of automation protocols, international reliability standards (e.g., $P_a < 10^{-7}$/yr) (Horowitz et al., 2019).

6. Survey/Interdiction, Safeguards, and SAR Applications

Uncrewed systems revolutionize wide-area search, safeguards inspection, decommissioning, and CBRN disaster response.

• Wide-area search and interdiction:
  • Aerial and ground platforms equipped with SiPM-based spectrometers and Compton imagers perform rapid source localization, leveraging beamforming and tomographic inversion to extrapolate activity off the survey path. The method

  $$\hat{s} = (G^T C_\epsilon^{-1} G)^{-1} G^T C_\epsilon^{-1} m$$

  provides source estimates with quantified confidence intervals, outperforming traditional direction-blind NaI(Tl) surveys in restricted or hazardous regions (Sinclair et al., 31 Jan 2026).

• Facility safeguards and inspection:

  • Autonomous UGVs perform integrated mapping, object ID recording, gamma sensing, and maintain semantic-rich information maps for IAEA compliance (Wang et al., 2019).
  • Heterogeneous teams (crawling bots, floor sweepers, manipulators) coordinate for comprehensive α/β/γ and swipe-based surveys in GNSS-denied environments, supporting time-integrated dose mapping, retreat protocols under high dose, and flexible teleoperation/autonomy switching (Pryor et al., 2024).
• SAR and emergency response:
  • UGVs/UGVs with tracked mobility and modular manipulators support rapid CBRN triage (radiation mapping, powder sampling, valve shut-off), minimizing operator exposure and providing accurate situational awareness (Schwaiger et al., 2024).
  • Aerial micro-vehicles (hexacopters, small quadrotors) employ multi-modal SLAM, NIR/depth sensors, and Bayesian localization to access confined spaces during decommissioning or post-incident reconnaissance (Mascarich et al., 2017, Thakur et al., 2019).

7. Future Directions and Technical Challenges

• Multi-robot learning and decentralized coordination: Robust, operator-agnostic relocalization, automated role reassignment, and task allocation are open problems under non-deterministic communication and field-of-view constraints (Vannini et al., 2 May 2025).
• Scalability and modularity: Dockerized software deployment, parameterized SLAM pipelines, and sensor/end-effector modularity are essential for broad applicability across site types and mission classes (Schwaiger et al., 2024).
• Robustness in degraded visual environments: NIR-LED flash, multispectral vision, and depth fallback enable resilient SLAM and mapping under darkness, dust, and occlusion typical of nuclearized environments (Mascarich et al., 2017).
• Limits and tradeoffs: Endurance, field of view, and real-time processing limits operational area and rate; grid cell size and SLAM update rates trade coverage against localization and radiation detection fidelity (Pryor et al., 2024).

Uncrewed mobile systems are thus a mainstay in advanced nuclear security, straddling C2, survey, interdiction, and SAR roles. Their deployment must be grounded in transparent technical standards, preserved human oversight, and robust international governance to balance the demonstrated operational benefits with accident, escalation, and stability risks (Horowitz et al., 2019, Pryor et al., 2024, Sinclair et al., 31 Jan 2026).