Cookie CodersFTC Team 39092
Programs / Auto-Learn-LimelightTracking
Autonomous · FTC Blocks // Student lesson

Auto-Learn-LimelightTracking

Blue Nectar: Center & Approach

Study Auto-Learn-LimelightTracking using the disabled Blue Nectar: Center & Approach excerpt, BlueNectarStudy.blk. It contains REAL motor-power commands but omits the side/rear wall guard and MUST NOT be run on the physical robot. Trace blue-contour filtering, centering, steering, loss handling, and stop reasons on paper. The estimated 3-inch lens gap is not a verified safe clearance.

The area-distance constant was calibrated at a longer distance, not verified close-up. The robot front can protrude beyond the lens. Under-250-ms results can be reread and counted more than once; the source does not prove distinct frames. Its encoder check detects little aggregate movement early in approach, not every individual frozen encoder or every later stall.

Start with a question

If an old picture still shows a blue ball, what stops a robot from driving toward a position that is no longer true?

By the end, you can…

  • Trace valid, pipeline, age, contour-area, and image-height acceptance tests.
  • Compute source range estimates, steering corrections, and the fixed encoder cap.
  • Explain centered-check counters, contour continuity, and immediate zero on loss.
  • Distinguish estimated lens-gap completion from encoder/time stops and physical clearance.
Start learningJump to CodeQuizDownload Blocks (.blk)

Before running: Ask your supervisor first. Keep Driver Station STOP ready.

Read · predict · explain

Build the idea, one block at a time

Read these steps before exploring the full program. The numbered blue comments in the diagram link back to the matching step.

01Recognize real outputs in a disabled study

Actual Blocks for this section
Blocks for Recognize real outputs in a disabled study
Blocks to notice
  • Direction
  • BRAKE
  • RUN_USING_ENCODER
  • leftPower / rightPower = 0
  • call applyPower

FL, BL, and BR are REVERSE; FR is FORWARD. All four use BRAKE and RUN_USING_ENCODER, and initial left/right requests are zero. applyPower really writes those requests to motors. DISABLED metadata prevents ordinary selection until changed, but it does not turn the source into a simulation. This excerpt omits the actual side/rear wall guard and must not run on the physical robot.

For example: A paper trace may contain leftPower 0.18. If this code were enabled and run, applyPower would send it to both left motors rather than merely display it.

Think first

Is this read-only like WallGuardStudy because both are disabled studies?

Show Answer

No. WallGuardStudy reports requests only; BlueNectarStudy includes real motor-power writes and must not be run on the robot.

02Identify the front blue-color camera

Actual Blocks for this section
Blocks for Identify the front blue-color camera
Blocks to notice
  • setPollRateHz 100
  • pipelineSwitch 0
  • Limelight start

The source selects blue color pipeline 0 on the forward-facing Limelight and requests 100 Hz polling. The separate rear Logitech camera is not used. Slot 0 must contain the expected blue-contour settings, unlike the AprilTag lessons' slot 0. A polling rate does not guarantee that every later read is a distinct image.

For example: AprilTag pipeline 0 and blue color pipeline 0 are different camera configurations, even though both are numbered 0.

Think first

Can a rear Logitech image satisfy readBlue's Limelight ColorResults tests?

Show Answer

No. This source reads the configured Limelight's result object and color contours.

03Preview detection without starting the motion phases

Actual Blocks for this section
Blocks for Preview detection without starting the motion phases
Blocks to notice
  • while opModeInInit
  • call readBlue
  • show tx / range / area
  • rangeInches - 1.4
  • sleep 50
  • waitForStart

INIT reads blue contours and reports angle, lens-to-center estimate, contour area, and estimated lens-to-surface gap = range - 1.4. Setup has already requested zero, and the motion phases begin only after waitForStart. A missing contour displays not detected. This preview and all subsequent motion logic are for source study only, not authorization to run the disabled excerpt.

For example: An estimated center range 7.1 inches gives a surface gap estimate 5.7 inches, using half of the assumed 2.8-inch ball diameter.

Think first

Does a 5.7-inch estimated lens gap prove a 5.7-inch robot-front clearance?

Show Answer

No. The robot front can extend beyond the lens, and the estimate is not close-range tape-verified.

04Gate centering with visible and centered checks

Actual Blocks for this section
Blocks for Gate centering with visible and centered checks
Blocks to notice
  • runtime < 6
  • centeredFrames < 5
  • seen AND ABS tx < 25
  • visibleFrames >= 5
  • ABS tx <= 2
  • sleep 25

Centering lasts at most 6 seconds. Only seen targets with |tx| strictly below 25 degrees advance visibleFrames; the first four accepted checks keep powers zero. Starting with check 5, |tx| <= 2 increments centeredFrames at zero; otherwise centeredFrames resets and positive tx requests left +0.12/right -0.12, with signs reversed for negative tx. Loss or |tx| >=25 sets both powers zero and resets both counters. Five centered checks are needed, but no image identity is checked, so these counters are not guaranteed distinct fresh frames.

For example: For a continually seen 0-degree target, checks 1-4 are visibility warmup; checks 5-9 increment centeredFrames to 5. At exactly 25 degrees this centering gate fails.

Think first

What happens if tx leaves the 2-degree band after centeredFrames reaches 3?

Show Answer

The found-but-off-center branch resets centeredFrames to 0 and selects a +/-0.12 turn; loss also resets visibleFrames.

05Keep an estimated lens target separate from clearance

Actual Blocks for this section
Blocks for Keep an estimated lens target separate from clearance
Blocks to notice
  • active AND centeredFrames >= 5 AND seen AND range > 4.4
  • tracking true
  • capture four encoder baselines
  • targetTicks = (range - 4.4)*115
  • runtime < 16

Approach enters only with sufficient centering, a seen contour, and estimated range above 4.4 inches. A 2.8-inch ball has radius 1.4, so 4.4 lens-to-center suggests a 3-inch lens-to-surface gap. The constant was calibrated at a longer distance; this close-up gap is not verified, and the robot front may project beyond the lens. The source fixes an encoder cap from the initial estimated range at 115 ticks/inch, remembers all four starting positions, enables contour continuity checks, and starts a 16-second maximum approach timer.

For example: Starting range 14.2 inches makes targetTicks = (14.2 - 4.4)*115 = 1127. That is a software cap from an estimate, not permission to travel toward a ball.

Think first

Does an initially centered target already at 4.4 inches enter the approach?

Show Answer

No. Entry requires range > 4.4; the else report says no target, not centered, or already inside the estimated gap.

06Read again and measure the largest wheel change

Actual Blocks for this section
Blocks for Read again and measure the largest wheel change
Blocks to notice
  • call readBlue
  • abs(CurrentPosition - startTicks)
  • MAX of four changes
  • sleep 25

Every approach loop calls readBlue and computes encoderTicks as the maximum absolute change among the four wheels, not their mean or sum. It later sleeps 25 ms, but processing adds time, so this is not an exact 40 Hz control guarantee. A repeated latest result may still pass the age test; refreshing the call does not prove a new frame. Displayed tx/range can retain old numbers after seen becomes false, so flags and commands matter.

For example: Changes 100, 105, 80, 0 ticks yield encoderTicks 105. A single frozen encoder is not exposed by this aggregate alone.

Think first

Would rereading the same 100-ms-old valid result necessarily fail readBlue?

Show Answer

No. It is still under 250 ms and there is no frame-deduplication check.

07Add and subtract a clipped correction

Actual Blocks for this section
Blocks for Add and subtract a clipped correction
Blocks to notice
  • seen AND ABS tx <= 25 AND range > 4.4
  • encoderTicks >= targetTicks
  • constrain 0.02*tx to [-0.18,0.18]
  • left = 0.18 + correction
  • right = 0.18 - correction

A seen target at |tx| <=25 and range >4.4 permits the approach decision. Reaching the fixed encoder cap selects zero and a safety-limit reason before another forward write. Otherwise correction = clip(0.02*tx,-0.18,+0.18); left adds it to 0.18 and right subtracts it. Commands range from 0 to 0.36, and the correction saturates at |tx| >=9 degrees. This forward differential steering does not strafe and is recalculated each loop.

For example: tx +5 gives correction +0.10, left 0.28, right 0.08. tx -12 clips to -0.18, giving left 0 and right 0.36.

Think first

Is tx exactly +25 degrees allowed by the approach steering gate?

Show Answer

Yes, if seen and range >4.4. The approach uses <=25, unlike centering's strict <25; correction is clipped to +0.18.

08Zero first, then distinguish goal, edge, and loss

Actual Blocks for this section
Blocks for Zero first, then distinguish goal, edge, and loss
Blocks to notice
  • leftPower / rightPower = 0
  • applyPower
  • if seen and range <= 4.4
  • edge abort
  • lossStartedAt
  • loss elapsed >= 0.35

Whenever the steering gate fails, both commands are zeroed immediately. If seen and estimated range <=4.4, the source sets done and an estimated-gap completion reason. Otherwise a seen target outside the angle gate aborts at the camera edge. An unseen target starts a loss timer and aborts after at least 0.35 seconds; it stays at zero throughout the wait. The close-range branch is tested before the edge branch, so a seen close estimate can report completion even if |tx| exceeds 25. That is another reason not to treat this study as validated safe behavior.

For example: At loss start, motors are zero, not allowed to coast forward under power for 0.35 seconds. Recovery into the valid guiding branch clears lossStartedAt to -1.

Think first

Does the loss timeout authorize driving blind for the grace interval?

Show Answer

No. Power is zero immediately; 0.35 seconds governs abandoning the approach, not when stopping begins.

09Separate safety-limit stops from estimated arrival

Actual Blocks for this section
Blocks for Separate safety-limit stops from estimated arrival
Blocks to notice
  • runtime >= 1 AND encoderTicks < 5
  • seen AND not done AND not abort
  • encoderTicks / 115 telemetry
  • done / abort / timeout cleanup

The early movement check aborts only when seen, not already done/aborted, time >=1 second, and the maximum absolute wheel change is <5 ticks. That detects very little reported movement across all wheels, not every single stuck encoder; it does not continuously detect a later stall once cumulative travel exceeds 5. Encoder-cap termination sets done but labels the result STOPPED, not gap success. The 16-second timer also stops without proving arrival. Final cleanup writes zero and stops the Limelight.

For example: At 1.2 seconds changes 4,3,0,2 have max 4 and trigger this check. Changes 100,100,100,0 have max 100 and do not expose the frozen fourth encoder through this test.

Think first

If encoderTicks reaches targetTicks while camera range is still 6 inches, is the estimated 3-inch lens gap achieved?

Show Answer

No. The source records an encoder safety-limit stop before the camera gap, even though it uses done to terminate the loop.

10Reset seen and require a recent result

Actual Blocks for this section
Blocks for Reset seen and require a recent result
Blocks to notice
  • seen false
  • getLatestResult
  • is not null
  • IsValid
  • Staleness < 250
  • PipelineIndex = 0
  • ColorResults length > 0

readBlue resets seen false and fetches the latest result. A non-null result may report age, global Ta, and contour count. Detection proceeds only with validity true (the source compares its text form with 'true'), age strictly under 250 ms, pipeline index 0, and a nonempty ColorResults list. Global whole-image Ta is displayed but not used for distance. Unlike the AprilTag samples, this has an explicit age gate; it still does not deduplicate frame identity.

For example: A valid pipeline-0 frame at 249 ms with qualifying contours can pass; the same conditions at exactly 250 ms cannot.

Think first

Can a valid 20-ms-old result from pipeline 1 guide this program?

Show Answer

No. PipelineIndex must equal 0 as well as passing age and validity.

11Select the largest eligible ball-height contour

Actual Blocks for this section
Blocks for Select the largest eligible ball-height contour
Blocks to notice
  • area = 0
  • for each ColorResult
  • TargetArea > area AND <= 0.75
  • 215 <= TargetYPixels <= 260
  • area >= 0.001
  • range = 2.84 / sqrt(area)

The source chooses the largest contour with area no more than 0.75 and Y pixels inclusively between 215 and 260. Only after scanning candidates does area >=0.001 set seen true. Equal-area candidates do not replace the earlier choice. tx comes from that contour's TargetXDegrees, not global tx. The distance model treats this contour area as the source's area fraction, using 2.84/sqrt(area); area units and the longer-distance scale need review. It is not a tape-verified close-range measurement and cannot distinguish every blue floor patch by color alone.

For example: Eligible area 0.04 gives 2.84/0.20 = 14.2 inches. A larger contour at Y 300 is rejected; one at Y 215 is within the band.

Think first

Should whole-image Ta replace the chosen contour's area in this range equation?

Show Answer

No. The source deliberately uses the selected contour's area; global Ta can combine unrelated blue regions.

12Reject continuity jumps without claiming identity

Actual Blocks for this section
Blocks for Reject continuity jumps without claiming identity
Blocks to notice
  • if tracking
  • ABS(tx-lastTx) > 12
  • area < 0.4*lastArea OR > 2.5*lastArea
  • seen false
  • update lastTx / lastArea only if seen

During approach, tracking rejects an angle jump greater than 12 degrees or an area outside [0.4,2.5] times the last accepted area. Boundary equality passes. Rejected candidates leave seen false and do not update the remembered accepted values, causing immediate zero in the caller. Selection chooses the largest eligible contour BEFORE this continuity test; it does not try another candidate if that largest one fails. These heuristics discourage target switching but do not prove the same physical ball.

For example: With lastArea 0.10, areas 0.04 and 0.25 pass the ratio boundary; 0.039 or 0.251 fail. An exact 12-degree jump passes, but 12.1 fails.

Think first

If the largest eligible contour fails continuity, does readBlue automatically fall back to a smaller matching contour?

Show Answer

No. It marks seen false after the chosen candidate fails; there is no fallback search.

13Locate the unguarded physical output stage

Actual Blocks for this section
Blocks for Locate the unguarded physical output stage
Blocks to notice
  • FL Power = leftPower
  • BL Power = leftPower
  • FR Power = rightPower
  • BR Power = rightPower

applyPower writes leftPower to both left motors and rightPower to both right motors. Their difference steers the chassis while moving forward; this stage contains no distance-sensor checks. Source notes say the complete guarded robot program checks walls before these writes, but that guard is omitted here and is not installed by this migration. Keep BlueNectarStudy disabled and never run it on the physical robot.

For example: left 0.28 and right 0.08 become FL/FR/BL/BR 0.28/0.08/0.28/0.08 in a paper trace. These are real output Blocks, not telemetry.

Think first

Could running WallGuardStudy separately make these four writes guarded?

Show Answer

No. It cannot intercept this OpMode, and the disabled blue excerpt must not be run on the robot.

Try it without a robot

Make the math make sense

eligible result = valid AND age < 250 ms AND pipeline = 0 AND contour count > 0
The source checks recent result eligibility, not distinct-image identity; all conditions must pass.
range estimate (in) = 2.84 / sqrt(chosen contour area)
Uses the source's contour-area scale, not global Ta. Quadrupling area halves the estimated range under this unverified close-up model.
estimated lens gap = range - 1.4 in; target center range = 3 + 1.4 = 4.4 in
The assumed 2.8-inch ball radius is 1.4 inches. This is lens-to-surface geometry, not robot-front clearance or a verified stop distance.
c = clip(0.02*tx,-0.18,+0.18); left = 0.18+c; right = 0.18-c
Positive tx adds on the left and subtracts on the right. Magnitudes saturate at 9 degrees; recalculate every eligible loop.
cap = (initial range - 4.4)*115 ticks; observed ticks = max(|four wheel changes|)
The cap is fixed at approach entry. Reaching it is a safety-limit stop, not proof of the estimated lens-gap goal.
continuity: |tx-lastTx| <= 12 AND 0.4*lastArea <= area <= 2.5*lastArea
Applies after choosing the largest eligible contour during tracking; it is a heuristic, not unique object identity.

Use a notebook or talk through your answer with a partner. You do not need a robot to predict what these blocks will do.

1. Filter a result and its contours

On paper use valid pipeline 0 results aged 249 and 250 ms. Candidate areas/Y pixels are 0.04/230, 0.16/300, 0.03/215, and 0.80/240. Which contour wins when the result passes? Calculate range and estimated lens gap. Explain why repeated accepted reads do not prove distinct frames.

2. Calculate steering and continuity

Compute correction and left/right powers for tx +5, -5, +12, and +25 degrees during eligible approach. With lastTx 0 and lastArea 0.10, test candidate angle/area pairs 12/0.04, 12.1/0.10, 0/0.25, and 0/0.251. Does the source try another contour after a failed continuity test?

3. Classify four stops

Start at estimated 14.2 inches and calculate targetTicks. Trace missing target at time 2.0 and still missing at 2.35; max encoder changes 4 ticks at 1.2 seconds; max reaches the cap with range still 6; and seen range reaches 4.4 before the cap. Name each reason and distinguish the source's done flag from verified success.

4. Explain the clearance problem

Draw a lens behind the robot's front edge and a 2.8-inch ball. Show why range 4.4 suggests a 3-inch lens gap but not a 3-inch robot gap. Write the evidence needed to validate close-up distance separately from the single longer-distance calibration; do not propose a physical run of this excerpt.

Full commented Blocks program

One source program, in one column. Calls connect any named helper routines; they are not separate programs. Click a numbered blue comment to return to its explanation.

100%
Open full Blocks diagram
Back to explanationsDownload Blocks (.blk)

Ask your supervisor before importing or running these Blocks. Viewing the program does not control the robot.

Section in context

Full Blocks program

The highlighted Blocks belong to this section. You can select another blue comment.

Open full Blocks diagram
Check your understanding

Your turn

Choose an answer, then check your thinking. You can change an answer and try again.

Your answers stay on this page and reset when you reload. No cookies, account, or saved answers.

0 / 9 correct · 0 / 9 answered

0
0
Question 1Which statement correctly describes BlueNectarStudy?

Question 2A valid pipeline-0 result has eligible contours and age exactly 250 ms. Does it set seen true?

Question 3Contours are area/Y 0.04/230, 0.16/300, and 0.03/215. Which supplies the range if the result is eligible?

Question 4The chosen area is 0.04. What range and lens-surface gap does the model estimate?

Question 5During eligible approach tx is -12 degrees. What left/right requests follow clipping?

Question 6The target becomes unseen. What happens during the first 0.35 seconds of loss?

Question 7Starting range is 14.2 inches. The maximum wheel change reaches 1127 ticks while range still reads 6 inches. What outcome is reported?

Question 8tracking is true, lastTx = 0 and lastArea = 0.10. Does tx = 12 and area = 0.04 pass the continuity test?

Question 9Wheel changes at 1.2 seconds are 100,100,100,0 ticks. Does the source's <5-tick movement check detect the frozen fourth encoder?

Reset clears every choice, explanation, and score.

For supervisors: 90-minute teaching plan & robot setup

Before the robot is used

Safety: DO NOT RUN BlueNectarStudy on the physical robot, including as a wheels-raised trial. Keep it disabled and do not deploy it. It contains actual motor commands, omits the side/rear wall guard, and cannot be guarded by WallGuardStudy in another OpMode. The 3-inch lens gap is an estimate from a longer-distance calibration, unverified close-up; the front can protrude beyond the lens. Use source inspection, paper traces, and independently supplied stationary evidence only.

Required hardware & dependencies

  • Four mecanum drive motors named front_left_drive, front_right_drive, back_left_drive, and back_right_drive; verify each configured name against its physical cable and wheel.
  • FTC Robot Controller, Driver Station, secure battery, and a notebook. Motor direction settings are specific to the source and mounting, not universal wiring instructions.
  • Source reference only: forward Limelight 3A named limelight with blue color pipeline 0. The rear Logitech camera is not an input to this excerpt.
  • Source reference only: four wheel encoders, assumed 115 ticks/inch, and a 2.8-inch blue ball. These constants are not validated by this lesson.
  • For learning: printed Blocks, notebook, calculator, and prerecorded or stationary measurement evidence that does not require executing this excerpt.

Source remains BlueNectarStudy.blk, DISABLED study-only. Native dependencies are Limelight 3A named limelight, a blue color pipeline 0, and four drive motors with encoder properties. No helper ZIP is provided. Review screenshots/Blocks and paper traces only; do not enable, deploy, or run this excerpt on the physical robot. The omitted wall guard and unverified close-range calibration cannot be supplied by another OpMode.

Robot setup checklist

  1. Open the disabled BlueNectarStudy.blk for inspection only. Verify the DISABLED flag, real applyPower writes, and absence of distance-sensor checks; do not change enablement.
  2. Review native Limelight result and ColorResults Blocks, pipeline 0 contents, contour-area units, Y-pixel band, and the documented single longer-distance calibration. No helper ZIP or deployed guarded program is supplied by this page.
  3. Mark lens position, assumed radius, robot-front offset, encoder scale, and unverified close-range behavior on a diagram. Do not infer a safe route from a 4.4-inch center estimate.
  4. Prepare synthetic paper results for age, angle, contour, continuity, loss, encoder-cap, and timeout cases. Treat counter values as loop checks unless distinct image evidence is separately provided.

90-minute teaching sequence

  1. 0-15 min

    Establish the no-run constraint and identify real outputs plus omitted guards.

    Look for: Students distinguish this study from read-only WallGuardStudy.

  2. 15-35 min

    Filter result eligibility and contour candidates, then calculate range and gap.

    Look for: Strict age boundary, inclusive height band, selected contour not global Ta, estimate not clearance.

  3. 35-55 min

    Trace visibility warmup, centered checks, steering saturation, and continuity.

    Look for: First possible fifth centered count at check 9, no frame deduplication, exact lock boundaries.

  4. 55-75 min

    Classify immediate loss zero, grace timeout, edge, encoder cap, movement check, and range branch.

    Look for: done is not a universal success flag; max ticks is not an individual-encoder health check.

  5. 75-90 min

    Quiz, clearance diagram, and exit ticket.

    Look for: No physical run proposed; close-up accuracy and robot-front clearance remain unverified.

Observation notebook

Paper notebook: frame eligibility, chosen contour, commands, and outcome
Result / contourseen / tracking matchEstimated range / gapLeft / right requestReason / caveat
Valid pipeline 0, age 249 ms, area 0.04, Y230true if continuity passes14.2 / 12.8 inAt tx +5: 0.28 / 0.08Model estimate, not measured clearance
Same conditions, age 250 msfalseStored numbers may remain0 / 0Immediate zero; loss timer may start
Cap 1127 ticks reached, range 6true6 / 4.6 in0 / 0STOPPED: encoder safety limit
Accepted contour, range <=4.4trueEstimated gap <=3 in0 / 0Estimated completion only; no close-up proof

Record source reason text separately from done/abort flags. Note repeated-result risk, lens/front offsets, and missing wall checks. These rows are paper traces, not physical trial instructions.

Troubleshooting · stop before investigating

No contour is accepted despite blue pixels.
On paper check validity, age <250, pipeline 0, nonempty contours, Y 215-260, area <=0.75, and chosen area >=0.001. Global Ta is not the range input.
Centering checks accumulate faster than processed images.
The source has no frame identity test; repeated under-250-ms results can increment counters. Do not call them verified distinct frames.
A larger blue patch causes seen to become false.
Selection picks the largest eligible contour before continuity rejection. Check >12-degree jump and 0.4-2.5 area ratio; there is no smaller-candidate fallback.
done is true but the reason says STOPPED.
The encoder cap uses done to terminate. Distinguish this reason from the seen-and-range<=4.4 estimated completion branch.
One encoder stays zero without the movement check failing.
MAX of four changes can exceed 5 even with a frozen wheel. This is not a per-wheel health or continuous stall detector.
The estimated lens gap seems safe for the robot front.
Reject that inference. The front may protrude beyond the lens, calibration is not close-up validated, and the excerpt must not be run.

Exit ticket & assessment

  1. State the no-run rule and contrast the two disabled study programs.
  2. Calculate clipped steering for tx -12 degrees.
  3. Explain why encoder-cap done is not success and why 4.4 inches is not verified robot clearance.

Assessment: Require correct result/contour gates, steering saturation, continuity boundaries, immediate zero-on-loss, and stop classification. A pass also requires the explicit physical no-run constraint, missing guard, repeated-frame limitation, aggregate-encoder limitation, and unverified lens/front clearance distinction.

Extension challenge: Analyze an independently collected stationary dataset to compare estimated range with tape range at several distances, especially close-up, and quantify lens/front offsets. Do not run, modify, or deploy this excerpt to collect it.

Teacher key · discuss after students predict

Contour and range practice

Age 249 passes, 250 fails. Area 0.04/Y230 wins; 0.16/Y300 and 0.80 exceed height/area limits, and 0.03/Y215 is smaller. Range = 14.2 inches and gap estimate = 12.8 inches.

Steering practice

tx +5 gives c +0.10 and 0.28/0.08; -5 gives -0.10 and 0.08/0.28; +12 and +25 clip to +0.18 and 0.36/0. Source centering rejects exactly 25, while approach steering accepts it.

Continuity practice

12/0.04 and 0/0.25 pass boundaries; 12.1/0.10 and 0/0.251 fail. A failed selected candidate sets seen false with no fallback or last-value update.

Stop practice

Cap is 1127 ticks. Loss commands zero at 2.0 and aborts when elapsed >=0.35; maximum 4 ticks at 1.2 seconds can trigger movement failure if seen and not done/abort. Cap at range 6 is a STOPPED reason. A seen range<=4.4 reports only estimated lens-gap completion.

Center and clearance limits

First four accepted center checks are visibility warmup, so five centered increments can finish at check 9; repeated frames are not excluded. Radius is 1.4 inches, but a robot front beyond the lens reduces real clearance. One longer-distance calibration does not verify this close-up stop.