Source-linked AI summary
Zephyron: Integrated Design and Analytical Evaluation of a Solar-Assisted Mobile Manipulator for Multimodal Environmental Reconnaissance and Distributed Visual Inference
Sabik Bin Sultan, Shafi Bin Sultan, Safwan Sadad
TL;DR
Environmental reconnaissance needs mobile platforms that preserve measurement context and return interpretable evidence under limited energy and communication. This paper develops a literature-informed Zephyron design and analytical framework, identifying distinct operating constraints and proposing quality-aware mission evaluation. Its conclusions remain design scenarios and require experimental validation before performance ratings can be assigned.
Problem
Environmental reconnaissance requires measurements linked to sampling procedure, instrument state, position, and time, while component specifications alone do not establish system performance.
Source: §1 · pp. 1–2Method
The paper establishes a literature-informed rover baseline, links subsystem calculations to mission decisions, and proposes quality-aware scheduling and evaluation.
Source: §1 · p. 2, §6.2 · p. 31, §10 · p. 47Results
The analysis identifies traction, manipulator reach, usable battery energy, sensor integration, calibration, and vision latency as distinct operating constraints.
Source: §9 · p. 45, §10 · p. 47Takeaways & Limitations
A matched comparison of continuous movement and stop-and-sample operation should assess measurement quality alongside coverage, energy, missed events, and interventions.
Source: §9 · p. 47The mission objective’s weights require a declared selection procedure, and the formulation claims no optimality or superiority over a fixed policy.
Source: §6.2 · p. 31
Abstract
from arXiv · showhide
Environmental reconnaissance needs mobile platforms that carry sensors, preserve measurement context, and return interpretable evidence under limited energy and communication. We present a literature-informed engineering design for Zephyron, a four-wheel rover with a front manipulator, environmental sensors, distributed computer vision, local recording, and a raised rear solar module. The design keeps the prototype layout but replaces unsupported numerical assumptions with an explicit component and geometry baseline. A reproducible search retrieved 5,000 records (4,858 unique) for screening, followed by targeted review of primary literature and manufacturer documentation. The baseline uses 165 mm wheels, a 12 kg mass budget, a 72 Wh battery-energy basis, and a 20 W photovoltaic module. With rolling-resistance coefficient 0.04, steady ascent of a 10 degree grade needs about 0.517 N m per wheel under equal load sharing. An illustrative 40 W motion load gives 1.44 h from 57.6 Wh usable energy, and a 25 percent driving duty gives 4.19 h without solar input; these are calculated scenarios, not measured performance. Sensor models show how integration time, calibration, temperature, and communication delay constrain interpretation, and a quality-aware stop-and-sample policy links these constraints to mission execution. Lightweight detectors, reference-based sensor learning, and executable data-integrity checks define a reproducible machine-learning evaluation pathway. The contribution is a traceable design and evaluation framework with editable 3D models, subsystem diagrams, and reproducible analytical data. Experimental validation is required before assigning payload, endurance, detection, or field-operating ratings.
1 Introduction
Zephyron is framed as a supervised environmental reconnaissance rover whose design preserves the photographed arrangement while making component choices and system claims traceable. The work addresses how to establish a coherent baseline, identify operating constraints, and schedule travel, sampling, and reporting.
1 Introduction
- Zephyron’s intended scope combines environmental screening, visual inspection, local recording, telemetry, and retrieval of small objects on a four-wheel rover with a forward manipulator and raised rear solar panel.Dimensions and ratings are new engineering selections, not measurements inferred from photographs.Source: §1 · pp. 1–2, Figure 1 · p. 2
- The paper identifies a gap between component specifications and system-level claims: motor torque, camera rate, or sensor accuracy alone do not establish field performance.Photographs establish visible arrangement, while renderings communicate proposed designs; neither is a calibrated field trial.Source: §1 · p. 2
- The design questions are how to establish a coherent component, mass, geometry, and power baseline; identify dominant constraints; and schedule travel, sampling, and reporting.The approach combines primary-source review, deterministic calculations, and a proposed experimental protocol.Source: §1 · p. 2
- The contribution is a traceable baseline, linked subsystem calculations, a feasibility-checking mission policy, and an editable visual and analytical package.The policy is proposed; its benefit requires testing against matched baselines.Source: §1 · p. 2
2 Research method and evidence traceability
The method separates observed evidence, verified specifications, assumptions, and calculated results to avoid assigning unsupported performance claims to the design. A reproducible literature search and focused review support an editable, machine-readable analytical workflow.
2.1 Source hierarchy and exclusion of unsupported quantities
- Photographs constrain relative arrangement but cannot establish absolute dimensions, component mass, or load ratings; selected component envelopes instead define a new geometry baseline.The evidence hierarchy separates observed features, verified specifications, external findings, design assumptions, and calculated results; numerical claims from the original concept paper are excluded.Source: §2.1 · p. 3
2.2 Literature search and focused review
- 5,000 retrieved records became 4,858 unique records, and a keyword screen flagged 1,630 for closer consideration; the corpus is metadata discovery, not full-text review.The focused review examined 25 relevant primary studies and was purposive, not a preregistered systematic review.Source: §2.2 · p. 3
2.3 Reproducible analytical workflow
- Deterministic analyses store their input grids and outputs in machine-readable tables with units and stated assumptions; curves represent sensitivity, uncertainty, or planning scenarios rather than experimental observations.The editable Blender model and subsystem diagrams communicate design intent but do not establish verified wiring, software, or integrated testing.Source: §2.3 · pp. 3–4
3 Related work
The related work supplies design and evaluation lessons while emphasizing that results remain tied to each study’s platform, environment, and test protocol. Across disaster response and environmental sensing, it highlights system integration, measurement context, and limits on transferring reported performance.
3.1 Disaster robotics as an integrated operational system
- Disaster-robotics studies emphasize whole-system access, communications, operator readiness, and task-specific control; their capabilities do not transfer automatically to Zephyron.The reviewed work also motivates testing environmental contamination, sight lines, tether behaviour, and recovery procedures.Source: §3.1 · pp. 4–5
3.2 Chemical sensing, mapping and the meaning of a measurement
- Gas measurements require sensor-state and calibration context: MOX response can involve warm-up, drift, slow response, and limited quantitative ground truth.The studies use chemical surrogates or simulations, so coloured maps do not independently validate hazardous-gas distributions.Source: §3.2 · pp. 5–6
- Gas-mapping routes can be assessed by information gained relative to travel and sensing costs, but object–gas associations remain hypotheses until a specified source-confirmation procedure is used.Remote-beam planning does not directly transfer to a point MOX sensor; planning abstractions must reflect sensor physics.Source: §3.2 · pp. 6–8
3.3 Radiation observation and uncertainty-aware motion
- Radiation studies motivate timing-aware pose association, acquisition planning, and explicit source and geometry assumptions, while surrogate experiments do not establish calibrated ionizing-radiation performance.Reported settings also bound transfer: one source-search experiment is barrier-free, and adaptive sampling studies report practical failures without general reliability estimates.Source: §3.3 · pp. 8–9
3.4 Water observation and sample traceability
- Water workflows support recording time, location, sensor status, and sample identity together; pH or threshold classification does not establish potability.Spatial interpolation remains dependent on sampling coverage, alignment, and calibration.Source: §3.4 · p. 9
3.5 Reproducible hardware and achievable autonomy
- Reproducible platforms support publishing hardware interfaces and test context, but external motion-capture positioning does not establish standalone outdoor localization.Localization methods also presuppose suitable sensors, calibration, and computation; benchmark accuracy cannot be assigned to an untested Zephyron package.Source: §3.5 · p. 9
- Service access, degraded-operation reporting, and repeatable task conditions are useful evaluation targets; manual module removal is not autonomous self-repair.A gripper alone does not demonstrate rescue extraction, chemical sampling, or force-controlled manipulation.Source: §3.5 · pp. 9–10
3.6 Evidence maturity across representative platforms
- The platform comparison groups systems by design-relevant role and avoids ranking accuracy or payload figures obtained under incompatible conditions.Table 1 is titled “Representative platforms and the limits of transfer.”Source: §3.6 · p. 10, Table 1 · p. 10
4 System integration and mechanical arrangement
Zephyron’s arrangement combines an accessible aluminum chassis, raised rear solar panel, and front manipulator, while treating rendered packaging as proposed rather than validated. The integration also specifies constrained steering modes and electrical and data-handling requirements.
4.1 Structure and service access
- The aluminum ladder structure carries wheel modules, payload mounts, battery, and solar supports, while access panels and restrained cable loops support inspection and articulation.The proposed rendering communicates the arrangement; its battery protection, enclosure details, and component packaging are not experimentally validated.Source: §4.1 · p. 10, Figure 7 · p. 11
- The elevated rear solar panel supports stationary collection but raises the center of mass and increases exposure to branch or doorway contact, requiring stability and clearance tests.The selected module measures 395 by 305 mm, and the support hinge height is near 720 mm.Source: §4.1 · p. 11
4.2 Coordinating traction and steering
- Grouped traction channels suit the specified symmetric turn, but general lateral motion requires reconsidering independent drive channels and documenting supported modes.Steering limits, actuator slew rate, wheel-speed saturation, and measured geometry constrain the kinematic command target.Source: §4.2 · pp. 11–12
4.3 Electrical and data interfaces
- The proposed power system separates solar charging, protected battery, motor supply, regulated logic, and switched mission loads, while final electrical ratings still require circuit design.The LiFePO4 pack requires chemistry-matched charging; conductor sizing, connector ratings, converter efficiency, inrush behavior, and thermal margins remain to be specified.Source: §4.3 · p. 12
- Acquisition records should preserve calibration, quality, operating state, and timestamps, with raw signals retained alongside converted outputs for later recalibration.Position records retain fix validity and antenna-to-sensor offset; camera model identity and frame timestamps remain distinct from detections.Source: §4.3 · pp. 12–13
5 Engineering models and calculated results
The engineering models establish a selected geometry and mass baseline, then calculate mobility and stability sensitivities under explicit assumptions. These are analytical estimates, not measured operating ratings.
5.1 Design scope and a reproducible baseline
- The selected 165 mm wheels, 538 mm track and 570 mm wheelbase define a new geometry baseline, while modeled clearances vary by component.The lowest central chassis component is approximately 191.1 mm above ground; folded-arm and fixed-probe clearances are approximately 96.1 mm and 141.3 mm.Source: §5.1 · p. 13
- The mass budget allocates 10 kg dry mass plus 2 kg mission payload, but these are targets rather than measured mass or demonstrated capacity.Grasped objects count toward the same 12 kg gross budget, and elevated components must be included in center-of-mass calculations.Source: §5.1 · p. 14
5.2 Mass allocation and geometric stability
- At illustrative center-of-mass heights of 0.20, 0.26 and 0.32 m, ideal tipping angles are approximately 53.4°, 46.0° and 40.1°.These are simplified geometric sensitivities, not operational slope ratings; tire, terrain and dynamic effects may reduce the margin.Source: §5.2 · p. 15
5.3 Mobility, traction, and geometric uncertainty
- The 0.864 m·s-1 peripheral speed is a no-load kinematic ceiling; the selected command cap is 0.35 m·s-1, and encoders do not directly measure ground speed under slip.Reduced turning models also omit skid-steer lateral slip and require terrain-specific parameters or a steering-geometry model for actively angled knuckles.Source: §5.3 · p. 16
- A 1% rolling-radius mismatch yields approximately 10.6° heading change and 0.92 m lateral deviation over 10 m in an idealized calculation.The calculation demonstrates sensitivity rather than predicting a field trajectory; effective radius can vary with wear, payload and surface slip.Source: §5.3 · p. 16
- At 12 kg, Crr=0.04 and 10° grade, equal sharing requires 0.517 N·m per wheel, below the 0.981 N·m continuous gearbox recommendation.The model estimates 25.08 N total force; the 10° grade is a proposed validation condition, not a gradeability rating.Source: §5.3 · pp. 16–17
- If only two contacts share the ascent demand, each requires approximately 1.035 N·m, exceeding the 0.981 N·m continuous gearbox recommendation.This sensitivity case shows that contact distribution can remove the apparent sizing margin; it is not a simulated suspension result.Source: §5.3 · pp. 16–17
5.4 Manipulator load budgeting
The manipulator is treated as a grasp demonstrator with explicit link and mass assumptions, not as a rated payload system. Its torque budget and sensitivity calculations identify loads and actuator capabilities that require later measurement.
5.4 Manipulator load budgeting
- The horizontal-arm model uses selected 0.146 m and 0.264 m links and assumed 0.12 kg, 0.16 kg, and 0.13 kg link and hand masses.These mass allocations are to be replaced after weighing; the grasp demonstrator must not inherit the chassis payload target.Source: §5.4 · p. 17
- For grasped masses of 0.15–0.25 kg, calculated shoulder gravity torque is 1.65–2.05 N·m, or 3.3–4.1 N·m with the assumed ×2 allowance.The allowance does not replace dynamic, shock, thermal, or structural assessment, and no 0.25 kg payload rating is asserted.Source: §5.4 · p. 18
- Initial grasp tests should use inert objects in a controlled fixture and report object geometry, jaw opening, measured grasp force, success definition, and release behavior.The next design step is torque calculation across a joint-angle grid using actual link centers of mass and acceleration profiles.Source: §5.4 · p. 18
- At 0.410 m extended reach, each added 50 g raises static shoulder moment by about 0.201 N·m, or the ×2 sizing requirement by 0.402 N·m.Adapters and cable attachments belong in the same load budget; bearing reactions, mounting compliance, cable drag, and transient acceleration remain separate loads.Source: §5.4 · p. 18
5.5 Energy architecture and mission duty cycle
The energy architecture combines a documented battery and solar module with explicit load, duty-cycle, and irradiance assumptions. Calculated scenarios and sensor procedures distinguish planning estimates from measured endurance or validated environmental readings.
5.5 Energy architecture and mission duty cycle
- The selected battery has a label-based nominal energy of 72 Wh, while the selected photovoltaic module is rated at 20 W; a matched solar charger and separately established protection and balancing functions are required.Actual delivered battery energy depends on discharge profile, cutoff, temperature, age, and cell state.Source: §5.5 · p. 20
5.6 Calculated energy scenarios
- At 25% drive duty, the 57.6 Wh usable-energy scenario gives 4.19 h in darkness and 9.22 h at constant 500 W·m-2 irradiance.The 35 W motion increment and weather are planning assumptions, not measured endurance or a forecast; additional arm, steering, lamp, and pump loads are excluded.Source: §5.5 · p. 20, §5.6 · p. 21
- A 20 h dark mission is infeasible under the stated 57.6 Wh usable-energy and 5 W auxiliary-load assumptions; six hours permits 13.1% driving in darkness or 34.6% at constant 500 W·m-2.The energy table’s values are steady-balance outputs, not endurance observations; sunny nondepletion still requires finite-capacity, time-varying analysis.Source: §5.5 · p. 20, Table 4 · p. 21, §5.6 · p. 22
5.5 Energy architecture and mission duty cycle
- An empty-to-full 72 Wh recharge has a 4.8 h energy-only lower bound at 15 W modeled full-sun input, with rover load disabled.Charge taper and changing irradiance lengthen the time; the separate temperature calculation must not double-count losses already included in combined derating.Source: §5.5 · p. 21, Figure 16 · p. 23
5.7 Environmental measurements and uncertainty
- The MQ-2 is a nonselective flammable-gas screening sensor with a specified 300–10000 ppm range and at least 48 hours of standard preheating.Its voltage is not a selective CO concentration or certified clearance result; calibration, interferents, and environmental conditions require characterization.Source: §5.7 · p. 22
- The SEN0244 specification is ±10% of its 1000 ppm full scale, equivalent to ±100 ppm, not ±10% of every indicated value.Its voltage conversion is a vendor example rather than a universal relation for floodwater; temperature reporting and field calibration remain essential.Source: §5.7 · p. 22
- The SEN0189 is described as qualitative and its response time does not establish sample settling or low-turbidity accuracy; quantitative claims require validation against a suitable nephelometer and standards.The proposed water run records calibration and reference information, uses blanks and check solutions, and flags values outside the validated range.Source: §5.7 · pp. 22, 24
5.8 Radiation counts and UV interpretation
Radiation interpretation is framed around counts accumulated over declared intervals, with uncertainty and background made explicit. UV reporting likewise requires spectral and contextual information, and the design claims no UV accuracy or exposure estimate.
5.8 Radiation counts and UV interpretation
- The first radiation output should be counts over a stated live interval, instrument status, and a background estimate, not a universal counts-to-dose conversion.Tube, high-voltage supply, pulse conditioning, energy response, and calibration must be specified before interpreting dose rate.Source: §5.8 · p. 24
- At a 1 count·s-1 expected signal, the planning scenario gives approximately 12.9% relative counting uncertainty at 60 s and 5.8% at 300 s.These are expected statistical limits before calibration or environmental contributions, not observed rover readings; low and zero counts require a declared Poisson interval procedure.Source: §5.8 · pp. 24–25, Figure 21 · p. 26
- A single uncalibrated photodiode channel cannot substitute for the spectral weighting used to estimate UVI, and exposure dose requires time integration and orientation/context.The dashboard should retain the estimated index, uncertainty or quality flag, timestamp, and sensor orientation; no UV accuracy or personal exposure estimate is claimed.Source: §5.8 · p. 25
5.9 Positioning and vision as distributed measurements
Position and vision outputs are treated as distributed measurements whose interpretation depends on reference frames, timing, bandwidth, and processing. The proposed evaluation keeps component benchmarks separate from rover measurements and avoids treating conditional latency sums as sustainable performance.
5.9 Positioning and vision as distributed measurements
- The NEO-6M’s 2.5 m horizontal CEP is a 50% containment measure under specified static conditions, not a 95% radial-error bound.Evaluation should document reference coordinates, antenna, satellite geometry, environment, outages, and statistical definition; antenna position must be distinguished from a deployed sampling tip.Source: §5.9 · p. 27
- For an articulated sampling point, antenna fixes alone cannot establish its location when attitude or joint state is unknown.The proposal stores antenna fixes, pose estimates, and mounting offsets separately, then compares stationary fixtures and repeated approaches to surveyed locations.Source: §5.9 · p. 27
- The camera subsystem sends compressed ESP32-CAM frames over Wi-Fi to an external computer for detection, keeping acquisition separate from inference.ESP32 memory and camera-driver constraints motivate recording frame size, buffers, firmware, and competing Wi-Fi/SD activity; published recording throughput is not rover performance.Source: §5.9 · pp. 27–28
- YOLOv4-tiny’s reported 443 FPS on an RTX 2080Ti is an external benchmark, not a Zephyron latency estimate or evidence of person-finding reliability in difficult scenes.The proposed evaluation reports edge hardware and end-to-end latency distributions as well as inference timing; no speedup or unchanged accuracy is asserted before measurement.Source: §5.9 · pp. 28–29
- With 5 FPS capture, 40,000-byte frames, a 2 Mbit·s-1 link, and the selected processing times, the queue-free latency sum is 360 ms, corresponding to 0.126 m travel at 0.35 m·s-1.It excludes braking, jitter, retransmission, detection failure, and operator response, and is not a safe stopping-distance estimate.Source: §5.9 · p. 28, Figure 24 · p. 29
- The ideal case requires at least 68.75% throttling/discard; this archived partial bottleneck upper bound excludes stage-concurrency and additional processing constraints and is not a throughput result.Source: §5.9 · p. 29
6 Mission design and quality-aware sampling
Zephyron’s proposed mission policy checks energy, actuator, and measurement-quality feasibility while adapting actions to each sensor’s evidence window. The section applies this approach to environmental sampling, mobility, retrieval, and stationary monitoring, with the illustrated tasks remaining proposed behaviors rather than demonstrated results.
6.1 A common mission state model
- Energy, actuator, and sensor-quality constraints must hold throughout a predicted trajectory; reserve violations reject an action before later charging can conceal them.Capacity saturation occurs before later discharge, and battery, converter, weather, and timing uncertainties belong in the feasibility margin.Source: §6.1 · p. 30
6.2 Precision and coverage tradeoffs
- Halving relative counting uncertainty requires four times as many expected counts, increasing energy use and reducing the locations visitable within a fixed mission period.For an assumed rate of one count per second, a 10 percent target requires 100 s and a 5 percent target requires 400 s.Source: §6.2 · p. 30
- A gas reading at the current GNSS fix may reflect earlier exposure; the response-length scale is not a calibrated plume-location error.Stop-and-sample operation, response identification, and explicit lag handling should be compared before displaying a continuous concentration map,.Source: §6.2 · pp. 30–31
- The proposed utility ranks candidate actions using information or coverage benefit and penalties for energy, time, and invalid measurements; its weights need a declared selection procedure.No optimality, reinforcement-learning performance, or superiority over a fixed policy is claimed.Source: §6.2 · p. 31
6.3 Water, gas, and radiation tasks
- Water sampling records raw and compensated signals with geotags and identifiers; gas screening records contextualized response, while radiation surveys retain integration time and statistical uncertainty.The water fixture does not establish waterproofing, and combustible-gas screening cannot establish that an atmosphere is safe for entry; the illustrated gas streamlines and radiation objects are not measurements.Source: §6.3 · pp. 31–32, Figure 25 · p. 31, Figure 26 · p. 32
6.4 Mobility, retrieval, and stationary monitoring
- Obstacle trials should report geometry and surface conditions alongside stopping behavior; ground clearance alone cannot establish traversability.Torque, tire contact, friction, mass distribution, and remaining wheel contacts also determine whether a wheel can lift onto an obstacle.Source: §6.4 · p. 32, Figure 28 · p. 33
- Retrieval evaluation separates approach, grasp, lift, retention, transport, and release; the rendered pose establishes neither rated capacity nor verified collision-free motion.Each stage must be evaluated against jaw geometry, friction, load, and actuator limits.Source: §6.4 · p. 33, Figure 29 · p. 33
- Stationary solar operation reduces movement duty, but energy estimates must account for panel-plane irradiance, temperature, and conversion losses rather than assume rated output throughout the day.UV monitoring also needs specified optical response and orientation; proximity to a solar module does not calibrate the sensor.Source: §6.4 · pp. 33–34, Figure 30 · p. 34
7 Vision inference, data integrity and validation methodology
This section presents a proposed, reproducible pathway for evaluating distributed vision, sensor learning, data integrity, and supervised operation, alongside guidance for interpreting uncertainty and structuring subsystem validation. It distinguishes software and analytical checks from unestablished detector accuracy, sensor performance, and field reliability, and cautions against treating repeated measurements as independent evidence.
7.1 Distributed camera and inference architecture; 7.2 Candidate detectors and training objective
- The proposed camera sends identified, timed frames to an external host for inference; compact SSD-MobileNet candidates and YOLOv4-tiny will be compared, not presumed demonstrated on Zephyron.The common starting input is one RGB frame resampled to 320 by 320 pixels; a two-dimensional box does not provide object distance or traversability. Candidate precedents are and.Source: §7.1 · p. 34, §7.2 · p. 36
7.3 Leakage control and calibrated interpretation; 7.6 Dataset construction and separation of development from evaluation
- Dependency-aware partitions must precede fitting, model selection, and calibration, with site-held-out testing preferred when independent sites permit it.The hash-based splitter uses selected proportions 55/15/15/15 but does not detect correlated observations or ensure balanced classes; a proposed water-conductivity forest requires independent reference-meter labels and temperature measurements.Source: §7.3 · pp. 36–37, §7.6 · p. 39
7.4 Executable checks and deployment acceptance
- Executable helpers test deterministic splitting, box matching, and conservative frame admission on constructed fixtures, while deployment acceptance requires matched floating-point and INT8 tests.Deployment measurements include inference distributions, peak memory, full-pipeline age, dropped frames, and host energy; no accuracy improvement or runtime advantage is asserted.Source: §7.4 · pp. 37–38
7.5 Timing, frame age and supervised operation
- Frame identity and age must remain visible to operators, and stale results must be excluded from current-scene assistance; recognition does not establish collision avoidance.The 0.35 m s−1 speed cap is a design constraint, not demonstrated safe speed, and the proposed freshness screening inequality depends on characterized braking and other conditions.Source: §7.5 · p. 39
7.7 Detection metrics and score interpretation
- Detection evaluation will report threshold-specific precision and recall separately from COCO-style average precision, with duplicate detections penalized and emitted-score reliability kept distinct from recall.Results will retain per-class outcomes and examine site, illumination, object size, and occlusion when subgroup evidence is sufficient; COCO evaluation follows documented ignore/crowd semantics.Source: §7.7 · p. 40
7.8 Quantization and matched computational comparisons
- Quantized and floating-point models will be compared on identical stored frames and in matched live pipelines, with operator fallback, resource use, and detection metrics reported.Recording-disabled and recording-enabled runs will isolate SD activity; transport losses, thermal state, and resource contention will be recorded because quantization alone does not guarantee speedup,.Source: §7.8 · p. 41
7.9 Uncertainty, comparators and the progression of evidence
- Uncertainty reporting separates measurement from sampling uncertainty, uses run- or site-level resampling for correlated frames, and stages evidence from feasibility through authorized field studies.No training accuracy, detection improvement, stopping performance, or field effectiveness is asserted before the proposed experiments are completed.Source: §7.9 · pp. 41–42
- 33 deterministic checks and 12 analytical regression tests passed, and 9,026 previously calculated numerical values were reproduced without mismatch under the comparison tolerance.The checks establish software behavior on specified fixtures and consistency of declared formulas; they do not establish detector accuracy, calibrated sensor performance, real-time deadlines, or field reliability.Source: §7.9 · p. 42
- At least 59 independent zero-miss trials are needed to bound the miss probability below 5% at one-sided 95% confidence for the tested condition distribution [59].Repeated frames or readings from one exposure are not necessarily independent trials.Source: Figure 35 · p. 43
- The mission decision cycle and distributed-vision scenarios are proposed illustrations, not detection results or field trials.The concepts depict camera observation with external inference processing and ground-station receipt of position, logging, and link-status information.Source: Figure 34 · p. 43, Figure 36 · p. 44
7.10 Validation matrix
- Validation should count runs, exposures, or independently prepared samples as appropriate, fix acceptance thresholds before held-out evaluation, and retain unsuccessful attempts in outcomes.Many frames or ADC samples from one event do not by themselves increase the number of independent trials; comparisons should use the same declared task definition.Source: §7.10 · p. 44, Table 6 · p. 44
8 Proximity supervision, illumination, and fault handling
Proximity input and illumination support supervised motion and inspection, but their readings and effects require explicit validity checks and matched testing. Fault handling must distinguish unavailable sensing from a clear path and evaluate stopping, steering, and arm motion as separate states.
Proximity supervision, illumination, and fault handling
- Missing or timed-out proximity readings are recorded as unavailable, not as a clear path; filtering delay counts toward the command-age budget.Acquisition time, validity, and configured range checks are required for an accepted reading.Source: §8 · p. 45
- Coverage and fault trials vary target position, test chassis stopping, steering transitions, and arm motion separately, and assess stop or hold responses to injected faults.Fault injection includes disconnected sensors, stale messages, and lost operator links.Source: §8 · p. 45
- Matched lamp-on and lamp-off recordings should retain exposure metadata because brighter images can still lose scene detail through glare, backscatter, clipping, or exposure transitions.Acceptance tests also record switched current, supply droop, and enclosure temperature.Source: §8 · p. 45
9 Discussion and engineering implications
The analysis identifies distinct mechanical, energy, and measurement constraints, and argues for quality-aware operation and matched experimental evaluation. Its compact models expose tradeoffs but limit transfer to field conditions.
9 Discussion and engineering implications
- Turning scrub, impacts, sinkage, and thermal accumulation remain outside the steady-grade torque scenario; arm reach amplifies gravity loads, and small-panel input cannot support unlimited operation under changing weather.These limits favor a supervised, intermittently stationary research platform with explicitly measured operating boundaries.Source: §9 · p. 45
- Longer integration can reduce counting uncertainty, but averaging cannot repair an uncalibrated turbidity conversion; gas-sensor drift and interferents can dominate electronic resolution.GNSS accuracy needs its statistical definition, while remote vision couples bandwidth and frame age to the distance traveled before results appear.Source: §9 · p. 45
- Restrained wiring, inspection covers, serviceable fasteners, and separated power distribution communicate construction intent but do not replace ingress testing, structural analysis, or manufacturing drawings.The recognizable layout constrains the redesign; those industrial details remain proposals rather than validation.Source: §9 · pp. 45, 47
- A matched comparison should test continuous movement against stop-and-sample operation using the same hardware, route, exposure conditions, and initial battery state.Compare valid-sample fraction, spatial coverage, uncertainty, mission energy, missed events, and operator interventions; energy savings matter only when task quality remains comparable.Source: §9 · p. 47
- Equal wheel-load sharing, constant rolling resistance, fixed PV derating, assumed auxiliary load, and simplified link masses expose tradeoffs but limit numerical transfer to field conditions.Future work should use measured rolling resistance and current maps, weighed mass properties, calibrated sensor models, time-varying irradiance, and uncertainty propagation that includes correlations and model discrepancy.Source: §9 · p. 47
10 Conclusion
The conclusion presents Zephyron as a reproducible design-and-analysis framework that links analytical constraints to mission decisions. It emphasizes that experimental work is still needed to establish integrated capability and measured performance.
10 Conclusion
- The baseline, mission illustrations, subsystem diagrams, and deterministic data make the design inspectable and reproducible.The proposed quality-aware policy connects traction, manipulator reach, usable battery energy, sensor integration, calibration, and distributed-vision latency to travel, sampling, and return-reserve decisions.Source: §10 · p. 47
- Detector candidates, conditional conductivity-regression, and executable frame-admission and matching checks define an assessable software methodology, while code checks establish implementation consistency rather than learned-model performance.Source: §10 · p. 47
- Experimental study must establish component integration, reference calibration, operating limits, and held-out task performance before measured capability claims.Source: §10 · p. 47