PaperlayerView exact revision on arXiv

Source-linked AI summary

A3P5 NEMESIS Integrated Rover Design for Environmental Reconnaissance and Robotic Sampling with Reproducible Mobility Analysis and an External Data Machine Learning Calibration Benchmark

Shafi Bin Sultan, Sabik Bin Sultan, Safwan Sadad

arXiv:2609.18245v1cs.RO

TL;DR

Environmental inspection requires a rover architecture that couples mobility, measurement and manipulation without treating missing measurements as demonstrated capability. This study reconstructs A3P5 NEMESIS, develops subsystem interfaces and evaluates engineering scenarios alongside a chronological public-data calibration benchmark. The benchmark reports a held-out CO RMSE of 0.502 mg/m3, while integrated rover performance remains to be established.

  • Problem

    The study asks how the existing physical arrangement can support a coherent environmental-inspection architecture without treating missing measurements as established performance.

    Source: §1 · p. 1

  • Method

    The study connects the photographed rover to a geometric reconstruction, proposed subsystem interfaces, reproducible engineering calculations and a chronological external-data calibration benchmark.

    Source: §1 · p. 1, §12 · p. 40

  • Results

    0.502 mg/m3 test RMSE was achieved by validation-selected ridge regression on the external sensor dataset, with a 95% daily-block interval of 0.435–0.569 mg/m3.

    Source: §12 · p. 40

  • Takeaways & Limitations

    The analysis identifies priority measurements and a testable route toward evaluating integrated inspection, while current support remains at the documented architecture and reproducible analytical-programme level.

    Source: §11.5 · p. 39, §12 · p. 40

    The public-data benchmark uses a different sensor array and deployment process, so it does not establish rover-specific accuracy or transfer to another device or site.

    Source: §10.5 · p. 36, §11.4 · p. 39

Abstract

from arXiv · show

A3P5 NEMESIS is a four-wheel rover intended to combine remote inspection, environmental observation and lightweight manipulation within one serviceable platform. This study develops a photo-constrained geometric reconstruction, a subsystem architecture and a reproducible analytical assessment while distinguishing physical prototype evidence from proposed functions. An exploratory search retrieved 5,000 bibliographic records across ten queries, yielding 4,897 distinct DOI records and 1,212 metadata candidates; selected primary studies and technical documents informed the design. The reconstructed configuration retains the carbon-pattern enclosure, independently steered wheel assemblies, folded manipulator, inclined camera mast and side sampling equipment. A declared 24 kg scenario predicts 3.28 newton-metres of gearbox-output torque per wheel on a 20-degree grade under equal load sharing; a separate static model shows how a 2 kg forward payload reduces the geometric front-tipping bound from 38.1 degrees to 32.7 degrees. These are design screens, not measured operating limits. A public-data calibration benchmark uses 7,344 eligible hourly observations, eight sensor/environmental predictors and chronological training, validation and test partitions. Validation-selected ridge regression achieves a held-out CO root-mean-square error of 0.502 milligrams per cubic metre, with a 95% daily-block bootstrap interval of 0.435-0.569 milligrams per cubic metre. This result concerns an external sensor array and cannot establish NEMESIS accuracy. The combined analysis identifies priority measurements, proposed control interfaces and mission-specific validation requirements. The contribution is a traceable engineering design study and evaluation framework for a prototype whose integrated field performance remains to be established.

1 Introduction

Environmental inspection couples sensor placement, measurement integrity, mechanical stability and operator interpretation, so component-level capability alone does not establish a system-level operating envelope. The paper connects the photographed rover to a proposed architecture and reproducible analyses while distinguishing established evidence from functions requiring future tests.

1 Introduction

  • Inspection requirements are coupled: wheel motion, motors, manipulator loading and communications delays can affect measurements, stability or stopping distance.The Blender reconstruction follows principal component locations in the photographs, while dimensions and material properties are not experimentally verified.Source: §1 · p. 1, Figure 1 · p. 2
  • A3P5 NEMESIS combines a carbon-pattern enclosure, four wheel modules, folded manipulator, inclined mast and side sampling equipment for remote inspection and environmental monitoring.Autonomous mapping, advanced planning and machine-learning interpretation remain development objectives, not outcomes established by the photographs.Source: §1 · p. 1
  • The central question is how the existing physical arrangement can support a coherent inspection architecture without treating missing measurements as established performance.The study separates photo-reconstructed geometry from metrology, assumed-input calculations from vehicle tests, and public-data modeling from rover calibration.Source: §1 · pp. 1–2
  • The paper documents the reconstruction, proposes subsystem interfaces, presents reproducible engineering calculations and evaluates a temporal calibration benchmark on public data.It does not introduce a new SLAM algorithm or claim a new sensing principle.Source: §1 · p. 2

2 Evidence, literature search and reproducibility

The study distinguishes photographed, described, modeled, literature-supported, analytical and public-dataset evidence, then documents an exploratory literature census and reproducible artifacts. The census describes a screened retrieval sample, not the research field as a whole.

2.1 Evidence hierarchy

  • No rover test log, measured mass distribution, motor torque curve, sensor co-location record or autonomy trajectory was available to establish integrated performance.The evidence framework distinguishes prototype presence from capability: a mounted camera, probe or steering actuator alone does not establish its accuracy or performance.Source: §2.1 · p. 3

2.2 Exploratory metadata census

  • Ten Crossref queries retrieved 5,000 records; DOI normalization left 4,897 unique records, including 1,291 with abstracts and 3,606 without.The documented screening rule prioritized 1,212 candidates, and a 150-record reading shortlist is a navigation aid rather than a claim of 150 critically appraised papers.Source: §2.2 · pp. 3–4
  • The census is exploratory, not a systematic review, and its counts do not estimate field-wide publication prevalence or growth.Relevance ordering, query caps, database coverage and missing abstracts constrain interpretation; Figure 2 describes the retrieved candidate sample.Source: Figure 2 · p. 3, §2.2 · p. 4

2.3 Reproducible artifacts and provenance

  • The supplementary package retains retrieval logs, screening scripts, candidate tables, engineering outputs, model artifacts and data hashes for reproducibility.Original scientific graphs and editable diagrams are accompanied by provenance records that distinguish rendered mission illustrations from prototype photographs and empirical figures.Source: §2.3 · p. 4

3 Related work

The related work frames NEMESIS as an integrated design whose mobility, manipulation and measurement requirements depend on mission-specific mechanisms and validation. Existing studies inform design choices but do not establish NEMESIS capability.

3.1 Scope of comparison

  • Mobility and measurement are coupled design problems, but combining familiar components does not establish NEMESIS novelty or quantitative reliability.The reviews address robotic gas-source localization and mapping, and dependable low-cost air sensing under cross-sensitivity, ambient change and drift.Source: §3.1 · pp. 4–5

3.2 Steering, terrain interaction, and manipulator stability

  • Four-wheel steering requires compatible wheel orientations and rolling velocities, while loading and wheel-contact geometry affect traction and tipover assessment.The visible steering modules do not establish feedback or tracking accuracy; NEMESIS evaluation requires identified masses, payload positions, wheel contacts and actuator limits.Source: §3.2 · p. 5

3.3 Integrated platforms and task-specific mechanisms

  • Rescue and telescopic-leg platforms illustrate mission-specific architectures, not evidence that NEMESIS can perform their tasks or match their mechanical performance.Karo and ResQbot address rescue tasks, while Mohamed et al. describe telescopic legs and omni-wheels; NEMESIS retains conventional tires and a different design scope.Source: §3.3 · p. 5

3.4 Air sensing and calibration

  • Low-cost gas channels should be treated as sensor responses pending calibration, and particulate corrections should be tested against NEMESIS reference observations.MQ135 has broad response, the checked MQ-7B operating sequence cannot be assigned to an unidentified MQ7 board, and BMP180 measures pressure and temperature rather than relative humidity. Particulate studies use differing sensor configurations and settings, so intake placement also needs testing against wheel dust, heat and water handling.Source: §3.4 · p. 6

3.5 Water sampling and the measurement interface

The related work treats water sampling, mapping and data analysis as interfaces that require defined procedures and deployment-relevant validation. The present design is an inspectable architecture, not a demonstrated capability or certification.

3.5 Water sampling and the measurement interface

  • NEMESIS water-sensing functions depend on accessible sampling positions and a defined probe interface; the layout does not establish reach, immersion depth, watertightness or contamination control.A meaningful comparison requires a full sampling procedure, including standards, rinsing, stabilization and repeated observations.Source: §3.5 · p. 7

3.6 Localization and environmental mapping

  • A mast location alone does not provide the localization, wind observations or measurement model needed to infer a concentration field.GADEN and VGR support repeatable simulation comparisons, but simulated environments differ from outdoor rover operation; localization stacks also require appropriate sensors, calibration and processing.Source: §3.6 · p. 7

3.7 Data analysis, external benchmarks, and validation

  • Public-dataset errors cannot be reported as NEMESIS calibration accuracy, and repeated mission readings should not automatically count as independent generalization examples.Preprocessing and model selection belong within training and validation data; final held-out missions, days, sites or instruments should remain untouched until evaluation.Source: §3.7 · pp. 7–8

3.8 Position of the present design

  • The design contribution is an inspectable rover architecture and staged assessment framework, not a certification or a claim of measured performance.The proposed sequence is to establish coordinated motion and load limits, characterize sensors and sampling interfaces, then evaluate localization and data interpretation under defined missions.Source: §3.8 · p. 8

4 Mechanical reconstruction and system architecture

The reconstruction preserves the photographed rover layout while making its geometric and wiring limitations explicit. A proposed functional partition separates low-level control, communications, and higher-level computation.

4.1 Geometry retained from the prototype

  • The four-view reconstruction preserves the enclosure, wheel modules, folded arm, inclined mast, and side sampling rack, while keeping the optional lift configuration separate.Photographs constrain arrangement, but perspective, occlusion, and the lack of a scale reference prevent a unique dimensional solution.Source: §4.1 · p. 8, Figure 6 · p. 9

4.2 Industrial detailing and cable management

  • The wiring model adds 39 cable or hose curves and 62 clamp assemblies, but checks only the reconstructed static pose, not full-travel clearance or fatigue life.Actual panel material, thickness, fastener spacing, and load paths are also needed for structural verification; the carbon-pattern appearance does not establish laminate properties.Source: §4.2 · p. 9, Figure 7 · p. 10

4.3 Functional partition

  • The proposed architecture assigns time-critical acquisition and actuator supervision to the low-level controller, networking and telemetry to the communication processor, and demanding computation to a companion processor.Commands should be bounded, validated requests, with sensor records carrying units, validity, calibration version, and acquisition time.Source: §4.3 · p. 10, Figure 8 · p. 11

5 Mobility, traction and steering coordination

The mobility analysis specifies compatible wheel commands, proposed feedback for tracking and traction monitoring, and analytical screens for grade demand. Its calculations rely on idealized kinematics and declared loading and friction assumptions, not measured vehicle performance.

5.1 Compatible wheel velocities

  • Compatible wheel commands follow from body translation and yaw rate under a locally planar, rigid-body, no-slip approximation.Steering travel, cable clearance, tire contact, and actuator torque still constrain physical point-turn feasibility.Source: §5.1 · p. 11, Figure 9 · p. 12, Figure 10 · p. 12
  • At 0.25 m/s and 0.140 m radius, ideal output-wheel speed is approximately 17.05 rpm, so gearing, encoder resolution, and continuous-operation heating need assessment.The analysis omits sinkage, bulldozing losses, and steering scrub, which matter on loose terrain.Source: §5.1 · p. 12, §5.3 · p. 13

5.2 Feedback, slip and implementation limits

  • The proposed feedback chain combines steering-angle and wheel-encoder feedback with independently estimated body motion to monitor tracking and traction consistency.Perpendicular wheel-velocity error is not a direct slip sensor when body velocity comes only from the same wheel encoders; this platform still requires its own identified model and validation.Source: §5.2 · pp. 12–13, Figure 11 · p. 13

5.3 Grade force and torque demand

  • 3.282 N⋅m per wheel is required on a 20° grade in the declared 24 kg, C_rr = 0.06, zero-acceleration case, assuming equal force sharing.This is gearbox-output torque at a 0.140 m wheel radius; the friction boundary is a simplified necessary condition, not evidence of loose-soil mobility or safe climbing.Source: §5.3 · p. 14, Figure 12 · p. 14

6 Manipulation, sampling and stability

The section treats manipulation, stability, and water access as proposed functions that require physical characterization. Its static model shows payload-sensitive geometric support bounds, while the mission illustrations establish only spatial compatibility in selected configurations.

6.1 Arm function and integration

  • The folded arm and deck location are retained, but its model is not a calibrated kinematic chain and does not establish payload capacity.Forward kinematics need measured joint axes, transforms, zero positions, and limits; payload capacity also requires link masses, gearbox ratings, bearing loads, and mounting stiffness.Source: §6.1 · p. 14, Figure 14 · p. 15, Figure 13 · p. 15

6.2 Payload-dependent support margin

  • 38.05° unloaded, 35.26° with 1 kg, and 32.66° with 2 kg are the modeled front-tipping bounds for the stated payload position.These are ideal geometric bounds, not operating limits; arm torque, gripper retention, and actual mass distribution remain unmeasured.Source: §6.2 · p. 16, Figure 16 · p. 17

6.3 Water access and sample logistics

  • The staged-vessel illustration places the pH electrode in liquid with its upper housing clear, but does not demonstrate automatic deployment or water-ingress protection.Sampling records should link containers to time, position estimate, mission identity, probe calibration, and acquisition state.Source: §6.3 · p. 17, Figure 17 · p. 17, Figure 18 · p. 18

7 Environmental sensing and measurement assurance

The sensing design treats environmental readings as traceable measurement chains rather than verified capabilities of the listed hardware. It specifies calibration, timing, and environmental checks while distinguishing analytical proposals from established performance.

7.1 Measurement scope and traceability

  • The listed sensor inventory is not a verified as-built bill of materials; concentration estimates require independent reference comparison and documented calibration and interface revisions.A numerical display alone does not establish analyte selectivity, traceability, or suitability for regulatory monitoring.Source: §7.1 · p. 18

7.2 Gas sensing, heater scheduling and electrical integration

  • MQ135 is broadly responsive rather than CO2-selective, and the MQ-7B heater schedule applies only after confirming the installed MQ7 variant.The proposed gas chain verifies sensor identity and heater operation, records phase and voltage, and rejects transition readings; wiring separation and motor-state logs help identify electrical artefacts.Source: §7.2 · pp. 18–19, Figure 19 · p. 19

7.3 Optical particles, humidity and sensor identity

  • Humidity effects observed with PMS1003 motivate quality control, not PMS7003 calibration; the proposed particulate chain also requires representative airflow and environmental covariates.The BMP180 measures pressure and temperature, not relative humidity; the correction described for PMS5003 is a candidate with unspecified A3P5 coefficients.Source: §7.3 · p. 19, Figure 20 · p. 20, Figure 21 · p. 20

7.4 Temporal response and environmental mapping

  • For a 15 s response time at 0.25 m/s, the modeled 90% response distance is 8.63 m; faster sampling cannot remove physical sensor lag.The response time requires experimental identification, and mapping trials should distinguish recovery transients from spatial gradients; Figure 23 shows scenario curves, not fitted response tests.Source: Figure 23 · p. 21, §7.4 · p. 22

7.5 Water probes, ultraviolet exposure and calibration uncertainty

  • At 25 °C, the ideal pH-conditioning increments are 0.0275122 and 0.00453951 pH/LSB for the respective ADC configurations; these are resolution values, not measurement accuracy.The water workflow includes conditioning and reference checks; turbidity and ultraviolet measurements still require module-specific calibration and documented uncertainty inputs.Source: §7.5 · pp. 22–24, Figure 25 · p. 22, Figure 26 · p. 23

8 Power, embedded control and communications

The proposed power and software architecture separates high-current loads from measurement paths and makes data validity and command authority explicit. Its energy estimates remain scenario calculations pending measured loads and representative mission logging.

8.1 Electrical power architecture

  • At 0.25 m/s, calculated endurance is 3.48 h on level ground and 1.72 h on a continuous 20° grade, under declared constant-load assumptions.The 10 Ah battery scenario yields 84.915 Wh after assumed derating; representative current and voltage logs and battery characterization are needed to replace the approximation.Source: §8.1 · pp. 24–25, Figure 28 · p. 24, Figure 29 · p. 25

8.2 Scheduling and data integrity

  • Acquisition records should retain timestamps, sensor identity, raw and calibrated values, validity flags, calibration version, battery state, and pose uncertainty.Separate actuator timing from slower sensing and network tasks so that web requests or sensor delays do not block bounded control updates.Source: §8.2 · p. 25, Table 3 · p. 25

8.3 Local networking and operator authority

  • An explicit command arbiter should reject stale or conflicting requests, while the local Wi-Fi dashboard remains a connectivity arrangement rather than a range guarantee.Video bandwidth should not indefinitely delay stop requests, and the dashboard should distinguish requested from active mode and show accepted-command age.Source: §8.3 · pp. 25–26, Figure 30 · p. 26

9 Navigation, mission concepts and protection

The proposed navigation and mission architecture links sensing, vehicle constraints, observation procedures, and protective states. Its illustrations and calculated stopping clearances define design and test requirements, not demonstrated autonomy or certified safety.

9.1 Proposed autonomy architecture

  • Autonomy requires pose estimation, obstacle representation, planning within vehicle limits, and control; mounting a camera alone establishes none of these capabilities.The proposed additions include wheel and steering feedback, an IMU, and range sensing; degraded localization should prompt a controlled stop or explicit recovery action.Source: §9.1 · p. 26, Figure 31 · p. 27, Figure 32 · p. 27

9.2 Mission-level behaviour

  • Mission completion requires a valid, contextually timed observation or confirmed manipulation outcome, not merely reaching a waypoint or issuing motor commands.Abort and recovery events should remain in the dataset; the reconnaissance illustration is not a rubble-traversal test.Source: §9.2 · pp. 27, 29, Figure 33 · p. 28, Figure 34 · p. 28

9.3 Stopping and fault handling

  • Calculated clearance rises from 0.430 m at 0.4 m/s to 1.450 m at 1.0 m/s under the stated delay, deceleration, and margin assumptions.These scenarios support braking trials, not a certified protective distance; fault injection should record detection time and actual stopping motion.Source: §9.3 · pp. 29–30, Figure 36 · p. 29, Figure 37 · p. 30

10 External-data machine-learning calibration benchmark

The benchmark evaluates CO calibration on eligible UCI Air Quality observations using chronological splits and training-only preprocessing. Validation-selected Ridge performs well on the frozen test set, but the historical external sensor array does not establish rover-specific accuracy.

10.1 Dataset, task definition and eligibility

  • The task used 7,344 eligible observations and eight contemporaneous sensor and environmental predictors, excluding other reference-gas columns and target lags.Unavailable targets were not imputed, and observations were neither interpolated nor concentration-filtered.Source: §10.1 · p. 30

10.2 Chronological evaluation and model selection

  • Chronological 70/15/15 partitions limit leakage from randomly mixing dependent observations, while 21 prespecified configurations were selected on validation data and frozen before testing.Preprocessing and fitting used training data only; there was no training-plus-validation refit or test-based retuning.Source: §10.2 · pp. 31–33

10.3 Metrics and uncertainty procedure

  • Daily-block bootstrap intervals resample observed calendar days and preserve within-day dependence, but omit across-day dependence, fitted-model uncertainty and missing-data uncertainty.The 48-day test block had 10–24 eligible hours per day; the resulting intervals describe frozen-model score variability, not future-reading prediction intervals.Source: §10.3 · p. 34

10.4 Held-out results and diagnostic interpretation

  • The selected forest’s test RMSE was 0.4871 mg m-3, slightly below Ridge, but the validation-locked choice remained Ridge; aggregate error does not establish performance during uncommon high-concentration events.Bootstrap interval overlap was not treated as a formal equivalence test.Source: §10.4 · p. 34
  • 0.5020 mg m-3 was Ridge’s test RMSE (95% daily-block interval 0.4353–0.5688), after its validation RMSE of 0.6853 beat random forest, boosting and the mean baseline.The validation RMSEs were 0.6900, 0.8202 and 1.5136 mg m-3, respectively; the test set contained 1,102 observations.Source: Table 4 · p. 34, §10.4 · p. 34, Figure 41 · p. 35

10.5 Computational cost, reproducibility and transfer limits

  • Ridge’s median warmed desktop single-row prediction time was 0.583 ms, but desktop timing does not establish embedded feasibility or transfer to NEMESIS’s different sensor array.The public-data benchmark also excludes unavailable targets and outages, which can introduce selection bias; rover transfer requires reference collocation and independent deployments.Source: §10.5 · p. 36, Figure 44 · p. 37

11 Integrated assessment and experimental validation

The integrated assessment treats scenario calculations as screens and identifies measurements needed to replace assumptions with an operating envelope. It proposes staged mechanical, sensing and mission validation while noting that measured integrated rover performance is absent.

11.1 Design tradeoffs exposed by the calculations

  • Mass and rolling resistance affect traction demand, centre-of-mass height affects stability, auxiliary power affects endurance, and response time affects interpretation of moving measurements.A taller mast may improve visibility while increasing structural loads and vibration sensitivity; increasing arm reach or payload may narrow the support margin.Source: §11.1 · p. 37, Figure 45 · p. 37, Table 5 · p. 38

11.2 Mechanical and control identification

  • Mechanical identification should replace visual assumptions with as-built mass, geometry and centre-of-mass measurements, then characterize motor and steering behavior under representative loads and trajectories.Repeat trials should document surfaces, slope, payload, battery state and failures, and distinguish path error from actuator saturation or slip.Source: §11.2 · p. 38

11.3 Environmental and mission validation

  • Environmental validation requires channel-appropriate references and procedures, while paired stationary observations and repeated traverses can help separate ambient changes from response lag and motion-associated effects.Synchronize reference and rover clocks, report position-assignment uncertainty, and leave unmeasured regions identified rather than using unqualified interpolation.Source: §11.3 · p. 38, Table 6 · p. 38

11.4 Threats to validity and limitations

  • The largest limitation is the absence of measured integrated rover performance; geometry, mass and energy scenarios remain assumed, and the public benchmark does not assess transfer to another device, site or gas-sensor model.The simple models omit coupled effects including unequal wheel loading, soft-soil interaction, suspension dynamics, motor thermal limits and actual arm-link movement.Source: §11.4 · p. 39

11.5 Research contribution and next experiments

  • The proposed next experiments connect controlled wheel-torque, centre-of-mass, power and co-location measurements to current uncertainties and saved analytical baselines.Subsequent field reports should retain provenance, state the tested configuration, describe exclusions and report negative outcomes.Source: §11.5 · p. 39

11.6 Reproducibility and project support

  • The research package includes the literature census, source registry, analytical scripts and tables, benchmark outputs, original graphs and Blender mission-study source.No newly collected human-participant or field environmental data are reported.Source: §11.6 · p. 39

12 Conclusion

The conclusion presents a traceable framework linking the photographed NEMESIS rover to its geometric reconstruction, subsystem interfaces and analytical results. Its scenario calculations and external-data benchmark identify candidate operating conditions and next measurements, not demonstrated integrated performance.

12 Conclusion

  • The framework connects the photographed rover arrangement to geometric reconstruction, proposed subsystem interfaces and analytical evaluation, with rendered missions requiring experimental confirmation.As-built parameter identification, reference co-location, fault-response tests and integrated field trials are identified as next steps.Source: §12 · p. 40
  • 3.28 N⋅m of output torque per wheel is required in the declared 24 kg, 20° grade scenario under equal load sharing, while a 2 kg forward payload lowers the modelled tipping bound from 38.1° to 32.7°.These are modelled screens, not measured operating limits.Source: §12 · p. 40
  • 0.502 mg/m3 was the external-data test RMSE for validation-selected Ridge, with a 95% daily-block bootstrap interval of 0.435–0.569 mg/m3; it does not establish NEMESIS accuracy.The benchmark used chronological evaluation and frozen test predictions.Source: §12 · p. 40

Related papers

Loading 2609.18245v1…