Integrated Design and Reproducible Feasibility Analysis of a Four-Wheel Rover with Packet-Commanded Imaging, LiDAR Ranging, and Learning-Assisted Target Recognition

Introduction

• Wheel speed, suspension, motor transients, and sensor interpretation interact, making rover feasibility an integration problem rather than a set of independent component choices.These interactions matter under competition, battery, and immediate-result constraints.Source: §1 · p. 1
• Under the 2027 NASA HERC remote-control high-school rules, autonomy means executing received task commands without manual task operations, not navigating the course autonomously.This scope allows a commercial radio-controlled drivetrain alongside a separately programmed task subsystem.Source: §1 · p. 1
• The proposed four-wheel architecture combines independent pivoting wheel carriers, four brushed geared motors, separate propulsion and task batteries, and a Raspberry Pi task system with camera, LiDAR, servo, display, and receive-only radio.The supplied material lacks a native SolidWorks assembly, calibrated mass properties, motor performance curves, and physical test logs, motivating the question of which limits can be derived and what must be measured.Source: §1 · pp. 1–2
• The study reconciles dimensional and mass estimates with mobility, suspension, power, and optical calculations; specifies a command-state architecture and hardware-independent verification suite; and evaluates learning-assisted classification on synthetic data.Experiments use generated images and mocked interfaces, with nominal and shifted scene distributions, rather than a completed rover.Source: §1 · p. 2

2.1. Literature retrieval and scope

• The literature search retained 3,453 unique records and 1,514 topic candidates, but was a documented metadata screen—not a systematic review or exhaustive coverage claim.Twenty-five primary references received closer assessment: 14 through selective full-text inspection and 11 at abstract or metadata level. Manufacturer documents and the handbook were treated separately as sources of specifications and task rules, not independent experimental evidence.Source: §2.1 · p. 2

2.2. Mobility, suspension, and printed wheels

• Ideal differential-drive equations are an initial command model, not universal turning parameters; terrain-specific identification and dynamic factors remain necessary.Skid steering involves lateral slip, and dynamic modeling also requires rolling resistance, terrain shear, actuator saturation, and controller behavior. The team-supplied CAD views are design renderings, not a fabricated or terrain-validated prototype.Source: §2.2 · pp. 2, 4, Figure 1 · p. 3
• Printed wheel properties cannot be assigned directly from prior studies because manufacturing conditions, processing, reinforcement, and material systems differ from the proposed PETG rim and TPU 95A tire.Prior work motivates component-level characterization, including TPU specimens and tires, PETG response studies, and a separate-rim/compliant-tire fabrication precedent; later multi-material and topology studies broaden the design space.Source: §2.2 · p. 4

2.3. Sensing and learning

• Single-beam camera–range registration requires an instrumented, multi-pose calibration adapted to limited range observations, beyond camera-intrinsic calibration alone.Planar-target calibration supports estimating camera intrinsics and distortion, while camera–range methods using richer observations underscore the need for overlapping geometry and independent calibration. Simple synthetic-board training does not establish transfer to outdoor images; the experiment is limited to visual classification of supplied board crops [21].Source: §2.3 · p. 4

2.4. Traceable design basis

• The adopted 421 mm camera height, 188 mm rim diameter, and 13.5 mm tire thickness follow the latest specification; inferred 540 mm track and 440 mm wheelbase are conditional estimates.The camera height excludes the antenna, so the complete excursion-ready envelope still needs measurement. The inferred track assumes symmetric tire placement and outer tire width; the wheelbase assumes wheel extrema establish overall length, rather than native-CAD measurements.Source: §2.4 · pp. 4–5
• 24.072 lb is below the handbook’s 60 lb limit but above its 20 lb scoring objective, requiring 1.847 kg (16.92%) removal to reach the objective.This is a mass-management target, not justification for removing load-bearing material without reassessing strength.Source: §2.4 · p. 5

3.1. Chassis, wheel carriers, and load paths

• The 6061-T6 chassis supports equipment and suspension, while each wheel carrier keeps its motor and spindle aligned and routes impact loads through two separated spindle bearings.Bolted modules permit replacement; batteries should be low and secured, and enclosure cooling and service access must not obstruct wheel travel or moving joints. The concept has eight springs and four dampers, and fabrication requires checking pivot alignment, fits, end stops, spring seating, cable radii, and full-travel clearances.Source: §3.1 · p. 5

3.2. Wheel manufacture and structural verification

• The PETG rim and TPU 95A tire require process-specific characterization; the hardness label alone does not establish tire stiffness, fatigue resistance, or tear strength.Manufacturing records should capture filament and print parameters, rim–tire bonding or interlock, and the interface retention check. Tests should characterize printed coupons, apply radial, lateral, and torsional loads to the rim/hub, and measure tire load–deflection, permanent set, traction, and repeated-use damage before correlating a finite-element model; no stress contour or safety factor is reported.Source: §3.2 · p. 6
• Spindle strength analysis requires actual support geometry, loads, and diameter; rendered-part thickness cannot supply missing inputs or capture local effects at threads, keyways, notches, and joints.Bearing reactions must use real support locations, since hub overhang can raise reactions beyond the simple wheel-load value. No stress contour or safety factor is presented because native geometry, boundary conditions, and material characterization were unavailable.Source: §3.2 · p. 6

3.3. Commercial differential drive and electrical connections

• The commercial drive uses one receiver signal per side to control paired reversible brushed ESCs, with only one ESC’s BEC positive connected to the receiver.The other BEC positive is isolated while signal and return remain connected; receiver failsafe should command neutral on both sides and be tested during radio loss under motor load. Proposed 20 A main and 5 A motor-branch fuses are provisional, and start, turn, stall, and climb currents require measurement with final wiring and load.Source: §3.3 · p. 7
• The jointly normalized throttle–steering mixer preserves left/right command ratio within range, but requires verification against transmitter behavior and motor orientation and does not map stick position to ground speed.Effective track in skid steering depends on slip, terrain, curvature, wheel loading, and suspension configuration; changing it models yaw sensitivity but does not compensate longitudinal slip. Pivot turns demand current and traction, and the drivetrain has no demonstrated encoder-based autonomous velocity controller.Source: §3.3 · p. 7

4.1. Wheel speed and the full-charge case

• At 215 mm diameter, 200 RPM corresponds to 8.10531 km/h, above the 8 km/h boundary; the threshold is 197.40148 RPM, so a nominal 200 RPM label does not demonstrate compliance.The assumed 190 RPM loaded-speed case is a scenario, not a measurement; light load and high voltage may define the maximum-speed case.Source: §4.1 · p. 9
• If a motor actually runs at 200 RPM at 12 V with approximately linear no-load voltage scaling, a fully charged 3S pack implies 210 RPM and 8.51057 km/h, an unverified endpoint estimate.PWM behavior, no-load specifications, tire deformation, battery sag, and terrain affect the result; use a provisional lower target and measure full-charge speed over a known distance, checking every permitted mode.Source: §4.1 · p. 9

4.2. Straight-climb lower bounds

• The aggregate traction inequality is necessary but not sufficient for each wheel, because grade-dependent load transfer and unequal contact loads can cause individual-wheel slip.The climb model assumes steady straight motion, equal torque sharing, and rolling resistance of 0.10; turning and transient losses are excluded.Source: §4.2 · p. 9
• At 20° and 2 km/h, calculated wheel mechanical power is 25.936 W; propulsion input is 43.227 W only under an illustrative 0.60 battery-to-wheel efficiency.The steady-climb torques are lower bounds, not procurement specifications; acceleration, turning, sinkage, wheel unloading, impacts, and drivetrain losses can increase demand. Continuous torque–speed–current data, rather than stall torque or ESC peak ratings, are needed to assess sustained operation.Source: §4.2 · p. 10

4.3. Static stability and changing contact geometry

• Under conditional geometry and an illustrative 0.200 m center-of-mass height, static tip-angle estimates are 53.47° lateral and 47.73° longitudinal, not approved operating slopes.Actual center of mass is unknown, and terrain contact, suspension, acceleration, tire deformation, and payload movement can reduce margin; losing a wheel changes the support polygon and requires recalculation.Source: §4.3 · p. 12
• Measure loaded corner weights with batteries and task hardware installed, then use controlled tilt tests or a validated mass-property model to locate the center of mass.Four equal shares would be 26.770 N each, but actual shares should be recorded; a center-of-mass-height sensitivity curve is more appropriate than a single maximum-slope claim based on an assumed height.Source: §4.3 · pp. 11–12

5.1. Local wheel rate and damping

• Spring behavior cannot be inferred by doubling one upright spring’s rate: attachment lever arms affect motion ratio, and local wheel damping is c_w = c_dρ_d^2.Significant linkage changes or preload may require a full force derivative with respect to wheel displacement.Source: §5.1 · p. 13
• Spring rates and sprung masses are uncertain inputs, so sensitivity sweeps guide component selection and measurement rather than predict ride traces.Wheels, carriers, and parts of the motor assembly may move differently from the chassis and should not automatically count as sprung mass.Source: §5.1 · p. 13

5.2. Integration checks and practical consequences

• A single loaded-corner test should measure wheel displacement, spring compression, force, and return behavior across intended travel to reveal motion ratios, preload, friction, binding, and end-stop behavior.Static testing can estimate the wheel-rate curve; damping requires dynamic measurement.Source: §5.2 · p. 13
• At full assembly, check wheel loads, ground clearance, and tire clearance through articulation and travel; because chassis roll and pitch alter sensor geometry, precise task measurements should occur from a settled position.Passive springs do not actively level the rover, and the drawings do not establish obstacle-height capability.Source: §5.2 · p. 14

6.1. Separate propulsion and task supplies

• The separate 3S propulsion and task packs total 79.92 Wh nominally, but their energy is not interchangeable without a suitable transfer path.The task pack feeds separately regulated branches; full-charge voltage is 12.6 V, so nominal voltage is not the maximum regulator input condition. The task rails and returns require correct wiring and hardware-revision checks.Source: §6.1 · p. 15

6.2. Low-side guard and shutdown behavior

• A pack-level low-side guard does not balance cells or provide orderly Pi shutdown, so the design requires per-cell alarms, verified voltage thresholds, shutdown reserve, and pre-run checks.The M148A switches the negative return; unintended ground connections can bypass that path, and the Pi must finish writing photographs and event records before power removal.Source: §6.2 · pp. 15–16

6.3. Illustrative current and runtime envelopes

• 222 W for propulsion and 97.68 W for tasks are the average battery-power budgets under an assumed 80% usable pack energy over a 12-minute mission.At 85% task conversion efficiency, the task budget corresponds to 83.03 W average output; these are arithmetic ceilings, not proof of continuous delivery or current capability.Source: §6.3 · p. 16
• 26.64 minutes at a constant 100 W propulsion draw and 39.85 minutes at a constant 25 W task output are modeled runtimes, not endurance tests.The simultaneous-load example assumes 10.5 V and 85% efficiency; the nominal 7.5 A versus 5.373 A difference is not a validated protection margin, because fuse curves, wiring, startup and stall currents, and regulator behavior matter.Source: §6.3 · p. 16

7.1. Task sequence and receiving architecture

• The receiver should enqueue validated decoded commands while operations execute, preserving order and repeated commands rather than discarding matching letters.The proposed 25-second service budget begins after dispatch and is not a hard packet-arrival-to-result deadline; queue waiting and cancellation acknowledgement are additional. The described chain uses VHF, RTL-SDR V3, demodulation, and Dire Wolf AX.25/APRS decoding, with final frequency and addressing following NASA’s released configuration.Source: §7.1 · p. 18

7.2. Fresh photographs and filter semantics

• Each D command should capture a fresh frame after pan settling, timestamp and uniquely store it; “no filter” means ordinary color, and optional recognition should use the unfiltered frame.A servo pulse alone does not establish angular accuracy, and usable travel must be calibrated against the permitted command sequence. The proposed channel-isolation filter is not necessarily the handbook’s intended interpretation or equivalent to an optical filter.Source: §7.2 · pp. 18–19

7.3. Pixel sampling and camera calibration

• A 12-inch feature at 50 ft spans approximately 70.96 pixels in the full 4608-pixel view and 29.57 pixels when scaled to 1920 pixels.This is an ideal sampling estimate, not a recognition guarantee; lens distortion, focus, exposure, motion blur, compression, cropping, and board pose affect the presented image. Calibration should use the selected capture mode and focus configuration across the usable image area.Source: Table 6 · p. 20, §7.3 · p. 20

7.4. Single-beam ranging and reference geometry

• At 50 ft, a 1° LiDAR pointing error displaces the ray 266.0 mm, while a 3° error displaces it 798.7 mm—beyond a centered 4 ft board’s half-width.The 0.5° beam-footprint calculation gives a nominal width of 133.0 mm, not a verified beam-profile measurement. If the ray still intersects the same flat plane, the slant-minus-normal difference is only 20.9 mm at 3°; averaging a wrong-surface return does not correct target association.Source: §7.4 · p. 21, Figure 18 · p. 23

8.1. State model and implementation boundary

• The executor uses WAITING, ACTIVE, and FINALIZED states; hardware errors or timeouts inhibit dependent actions until deliberate recovery, while finalization preserves partial evidence.Recovery is exposed as a required flag, but physical recovery and automatic flag clearing are not implemented.Source: §8.1 · p. 24
• The executor accepts validated tokens rather than defining the radio protocol, and its illustrative ±90° pan limits do not establish adequate mechanical travel for competition commands.Out-of-range turns are rejected and logged; deployed adapters must follow NASA’s released specification, persist required records, and define bounded safe hardware behavior.Source: §8.1 · p. 24

8.2. Executed checks and deployment tests

• 30 mocked-hardware checks cover command ordering, invalid inputs, concurrency, timeouts, cancellation, and sequence finalization.The checks also reject Boolean distances and prevent photo-adapter results from overwriting executor status.Source: §8.2 · p. 25
• The checks establish reproducible unit-level failure behavior, not RF, servo, camera, storage, or LiDAR performance; commissioning should progress from supported stationary operation to combined drive and task tests.The next stage replaces mocks with device adapters, replays sequences under fault injection, and verifies fresh results after an unknown sequence without external editing or a pit return.Source: §8.2 · p. 25

9.1. Dataset and experimental separation

• The benchmark used 2,400 generated 64 × 64 crops in 600 scene groups, with whole groups held within one split; shifted tests changed backgrounds and appearance conditions but were not real-world domain tests.The package preserves sampled parameters and image hashes for reproduction, but changing several nuisance factors together prevents attributing performance loss to any one corruption.Source: §9.1 · pp. 25–26

9.2. Features, classifiers, and model selection

• The comparison covered a color-and-geometry baseline, a 62-feature hue–saturation–value histogram classifier, and a linear SVM adding 1,764 HOG features; validation macro-F1 selected C from three values.Accuracy and macro-F1 intervals used 1,000 bootstrap resamples of complete four-image scene groups and reflect scene-sampling variation conditional on fitted models, not training randomness or physical-camera uncertainty.Source: §9.2 · pp. 26–28

9.3. Results and the value of a negative comparison

• The geometric baseline outperformed both learned models; HOG added no nominal accuracy and reduced shifted accuracy by 13.25 percentage points, while learned models rejected only 34 of 100 nominal no-target examples.The 95% accuracy intervals were 91.00–98.25% for the geometric baseline and 79.50–84.75% for the HOG-plus-color model on nominal data, and 65.50–78.00% and 43.49–52.75% under the shifted generator. The HOG-plus-color artifact was 98,188 bytes; median feature-extraction and one-image prediction time was 3.466 ms on a shared x86-64 Xeon host, not a Raspberry Pi.Source: Table 9 · p. 28, §9.3 · p. 28, Figure 21 · p. 29

9.4. Operational boundary and next experiment

• The crop classifier does not locate boards, produce calibrated probabilities, issue commands, or control motors; real-camera testing must separately assess localization and classification across acquisition sessions and target conditions.Fixed color–shape associations, finite generated negatives, and ideal supplied crops leave real-marker appearance and full-frame localization unvalidated; target-hardware measurements are needed to establish command-budget latency.Source: §9.4 · p. 30

10.1. Verification matrix

• Verification prioritizes full-charge speed, loaded-motor climbing, suspension geometry, voltage under simultaneous loads, and camera–LiDAR target checks; propulsion and energy remain scenario-only, while the synthetic experiment is complete.Tests should use a fixed hardware and software revision; the matrix avoids invented acceptance tolerances.Source: §10.1 · p. 30

10.2. Build budget and design-change control

• BDT 183,094.30 is the planning component budget: BDT 72,240.00 for chassis and drive items plus BDT 110,854.30 for task items.These are not purchase receipts or a complete lifecycle estimate, and exclude unrelated travel and event expenses.Source: §10.2 · p. 31
• A design change should update the component list, mass estimate, wiring, analysis inputs, and test record together because component substitutions alter multiple performance assumptions.For example, changing a wheel affects speed, torque leverage, clearance, and print validation.Source: §10.2 · p. 31
• Removing 1.847 kg would meet the 20 lb scoring objective, but unverified thinning or shortening could compromise reliability; the study prescribes no unvalidated lightening pattern.The recommended sequence is to identify nonstructural mass, quantify each change’s consequences, then recheck strength and operation.Source: §10.2 · p. 31

10.3. What the study establishes

• The study establishes geometry and scenario consistency, mocked reference-software behavior, and reproducible performance for three classifiers on synthetic crops, not physical task completion.Major limitations include no measured component performance, physical camera data, calibrated center of mass, or full-system trials; the model is quasi-static, printed-material properties are uncharacterized, and synthetic vision lacks simulation-to-real validation.Source: §10.3 · p. 31
• Future experimental work should report measured methods, uncertainty budgets, repeated trials, and failure cases rather than relabeling current calculations as experimental results.Literature can provide methods and comparison context but cannot supply missing observations for this rover.Source: §10.3 · p. 31

11. Reproducibility and Artifact Provenance

• The source package includes analysis scripts and CSVs, a reference executor, synthetic generator, trained classifiers, predictions, figure sources, and editable design assets.Analytical scripts regenerate six multipanel figure groups and check unit identities, wheel-dimension closure, mass conversion, and equivalent force–velocity versus torque–speed power; the software suite regenerates mock-adapter results.Source: §11 · p. 31
• ML verification reloads saved models, reproduces test predictions, recomputes metrics, and checks selected generated-image hashes.The supplied commands run the analysis, executor checks, and vision experiment after installing the listed ML requirements.Source: §11 · p. 32
• CAD screenshots remain design evidence, while Blender scenes are procedural approximations; task-scene renders are separate from the synthetic classifier dataset.Canva diagrams remain editable, reproduced comparison figures are limited to two identified CC BY 4.0 sources with attribution, and copyrighted full-text downloads are not redistributed.Source: §11 · p. 32
• Checksums support reproducibility and version identification, not physical correctness; the manuscript and data distinguish calculated, software-tested, and proposed elements.The literature log also preserves 20 queries, metadata deduplication, selected references, access levels, and claim limits.Source: §11 · p. 32

• At 10.919 kg, the estimated mass exceeds the 20 lb scoring objective by 1.847 kg; 200 RPM implies 8.105 km/h, and a 20° climb requires at least 1.255 N·m per wheel.The climb estimate assumes the declared resistance and equal torque sharing, before turning, transient, and other unmodeled demands; nominal motor speed does not establish compliance with the speed limit.Source: §12 · p. 32
• At 50 ft, a 1° aiming error shifts the ranging ray by approximately 266 mm, making target association, calibration, and the distance reference central integration requirements.Separate battery and regulator branches, correct low-side return routing, bounded command execution, fresh data, and visible failure records provide an implementation basis; 30 mock-adapter checks verify reference-software behavior without claiming hardware execution. Physical mobility, electrical, sensing, and complete-sequence measurements remain necessary to establish actual rover performance.Source: §12 · p. 32
• HSV-and-geometry rules outperform both linear SVM models in the synthetic experiment, while adding HOG reduces shifted accuracy rather than improving it.Learning-assisted recognition remains an extension requiring evaluation with real-camera data.Source: §12 · p. 32