Skip to content
Tech Interview Prep

Top 100 Robotics Engineer Practice MCQs

Every question on one page: open one, choose an answer, and see why at once. Signed in, your latest answer to each question counts, and you can answer again after 24 hours.

Curated: · Written: · Reviewed:

Reviewed 100Review pending 0

Answers to reviewed questions count toward this role and the overall leaderboard; your latest answer to each one is the one that counts.

100 questions

MCQ-3A pose filter fuses four orientation estimates each cycle by averaging their quaternion components and normalizing the result. After a sensor swap, the fused orientation intermittently flips by about 180° while the robot is stationary. The four input quaternions (w, x, y, z) logged for one bad cycle are q1 = (0.707, 0.707, 0, 0), q2 = (0.707, 0.707, 0, 0), q3 = (-0.700, -0.714, 0, 0), q4 = (-0.700, -0.714, 0, 0). All four describe the same physical orientation. What is the cause of the flip, and what is the correct fix?
MCQ-1A fixed camera looks straight down at a pick table. Hand-eye calibration gives the camera-to-base transform as a pure rotation of 90° about the base z-axis: a point (x, y, z) in the camera frame sits at (−y, x, z) in the base frame. Two parts are detected and picked: | Part | Detected in camera frame (mm) | Placed in base frame (mm) | | --- | --- | --- | | A | (200, 0, 0) | (0, −200, 0) | | B | (0, 200, 0) | (200, 0, 0) | Hand-eye calibration has been re-run twice and returns the same transform each time. Which explanation fits both placements, and what is the fix?
MCQ-88Picks on a cell land 90° out about one axis. The pick path converts the camera's poses into the cell convention before commanding the arm; the calibration-verification script reads the same raw poses and compares them to the taught pose without converting, so it reports clean. A diagnostics path added last quarter reads raw poses too, and the team adds a new consumer of raw poses most sprints. The arm, the PLC and the safety envelope all work in the cell convention. Which change addresses the cause?
MCQ-4A six-axis arm is being commissioned. Its URDF was generated from CAD. A laser tracker measures the TCP at commanded poses and the result is compared with the URDF's forward kinematics. The error is repeatable to 0.1 mm and does not change when a pose is approached from the opposite direction. | Pose | Joint 2 angle | TCP distance from joint 2's axis | Error magnitude | Error direction | | --- | --- | --- | --- | --- | | A | 0° | 900 mm | 3.9 mm | perpendicular to the joint 2–TCP line, in the arm plane | | B | 60° | 900 mm | 3.9 mm | perpendicular to the joint 2–TCP line, in the arm plane | | C | folded | 120 mm | 0.52 mm | perpendicular to the joint 2–TCP line, in the arm plane | Which single correction to the URDF accounts for all three measurements?
MCQ-14A 6-axis arm's controller computes the joint-5 gravity torque as m·g·d from the payload parameters. The CAD model lists the tool at 3.2 kg with its centre of mass 60 mm from the flange. After the tool is fitted with its cabling and a workpiece, the controller's identification routine reports 4.1 kg at 104 mm. At a full-extension pose the joint-5 position error is 0.6 mm; near-flange poses hold within 0.05 mm. Which conclusion does the measurement support?
MCQ-94A 6-axis arm is jogged from a pendant. The pendant's frame selector was last set to 'tool' in a previous session and its indicator sits on a second screen. The wrist is rolled 90° about the tool Z axis relative to base. An operator intending to raise the tool presses a 40 mm +Z jog and the TCP moves 40 mm along base +Y into a fixture. Which conclusion is correct?
MCQ-83A 6-axis arm's controller composes each new tool pose by multiplying the previous pose matrix by a small incremental transform, at 1 kHz. On the bench, after 400,000 updates: about 6.7 minutes of continuous motion: the accumulated rotation block has determinant 1.0012 and orthonormality residual max|RᵀR − I| = 4e-4. A commanded 2.000 m reach from the base now measures 2.0008 m, and the error grows with reach. Which conclusion do these numbers support?
MCQ-10A 6-axis robot carries a 180 mm tool. Its TCP was taught with four poses spanning 22° of wrist rotation, and the controller reports a pivot residual of 0.8 mm. Parts land correctly when the wrist stays near the teach orientation, but once the wrist rotates through large angles the tip lands 10.7 mm off. What does the evidence show, and what should be done?
MCQ-12A six-axis arm drills holes from CAD-generated offline programs. Taught points repeat to 0.02 mm, but the CAD programs land 0.8–1.0 mm off nominal at the far corners of the envelope. The vendor datasheet gives ±0.02 mm repeatability and ±0.9 mm absolute accuracy; kinematic calibration is quoted at ±0.25 mm. The cell must hold ±0.3 mm hole position. What is the right next step?
MCQ-11An eye-in-hand cell is calibrated before commissioning. The 40-pose set was collected by translating the camera across a 300 mm table region; the largest rotation between any two poses is 4°. The solver returns a 0.4 px RMS reprojection residual. Picks land within 0.5 mm at the 400 mm teach distance, but at 700 mm the tool misses by about 4 mm, always in the same lateral direction whether the target is left, centre or right of the table. What explains the miss and what should be done?
MCQ-28A direct-drive shoulder joint on a 1.4 m arm carries a 17-bit absolute encoder (131,072 counts/rev). Tool placement error measures 0.31 mm RMS. The measured error budget: thermal drift 0.24 mm, link compliance under payload 0.14 mm, kinematic calibration residual 0.11 mm, encoder quantisation 0.067 mm. A 23-bit encoder (8,388,608 counts/rev) is offered as a drop-in swap. What does the budget say the swap buys?
MCQ-63A six-axis arm loads a machining cell to a ±0.15 mm placement tolerance. The engineer logs placement error against structure temperature across two shifts. Warm start: 0.02 mm at ΔT = 1 K (07:00), 0.11 mm at ΔT = 6 K (08:00), 0.18 mm at ΔT = 10 K (09:00), 0.27 mm at ΔT = 15 K (11:00). Cold Monday: the same 0.18 mm appears at ΔT = 10 K, 40 minutes earlier in the shift. First-off inspection at 07:00 passes on both days. Which change should the engineer make?
MCQ-89A cell estimates a part's pose from a single camera before an insertion step. Measured per-axis uncertainty at the working distance is lateral 0.9 mm, vertical 1.0 mm, depth 12 mm. The insertion needs depth within ±2 mm; lateral and vertical tolerances are ±3 mm. The team has budget for one more camera and two mounting points: a bracket 120 mm from the first, or the far end of the frame 900 mm away. Bench measurements give per-axis uncertainty: | Axis | One camera | Two, 120 mm apart | Two, 900 mm apart | | --- | --- | --- | --- | | lateral | 0.9 mm | 0.7 mm | 0.6 mm | | vertical | 1.0 mm | 0.8 mm | 0.7 mm | | depth | 12 mm | 11 mm | 1.4 mm | Which placement should the team choose?
MCQ-90A cell uses a single 60 mm fiducial marker on the part fixture, viewed by a fixed camera at 2.4 m standoff, so the marker subtends about 22 px. The marker is detected in every frame with high confidence, but the reported pose jumps between two orientations on roughly one frame in eight, and the robot's approach direction reverses on those frames. What is the correct remediation?
MCQ-32A stereo depth camera in a bin-picking cell ran for a month at a 0.5 m mount height, the working distance at which its datasheet quotes 2 mm depth error, and picks were clean. A conveyor change raised the camera to 1.4 m; the parts, lighting, grasp planner and 5 mm grasp tolerance are unchanged, and pick failures went from under 1 percent to 12 percent. Which conclusion fits these data?
MCQ-33A cell places parts anywhere across the camera's field of view at working distances from 300 mm to 700 mm; every part the feeder presents must be placed, and the placement tolerance is 0.5 mm. The calibration set was captured with the target filling the central third of the frame at 500 mm. Its reprojection residual is 0.2 px, but a check at the frame corner shows 3.1 px of undistortion error, which is 2.4 mm at the working distance. Which approach should the team adopt?
MCQ-34A cell inspects 0.5 mm defects with a camera whose pixel footprint is 50 µm/px, so a defect spans 10 px. Recall held at 0.97 while the line ran at 0.6 m/s with a 250 µs exposure. After the line was raised to 1.8 m/s with the same exposure, recall fell to 0.68. The line must keep running at 1.8 m/s, and the lighting can be driven 3× brighter. On this cell, recall stays above 0.95 only while blur is at or below 30% of the smallest feature. What should the team do?
MCQ-27A gearbox accelerometer channel is sampled at 200 Hz with no analogue anti-alias filter ahead of the ADC. The gear mesh frequency is known to be 283 Hz. The recorded spectrum shows a clean peak at 83 Hz that matches no shaft, gear or bearing rate. What is the 83 Hz peak, and what is the correct next step?
MCQ-26A type-K thermocouple on a jacketed process vessel spans 0–400 °C and its control band is 150–180 °C. Mid-batch the channel reads 21.4 °C and has not moved for 90 minutes, while the heater runs at 100% duty and an independent RTD on the jacket climbs to 176 °C. The range check (0–400 °C), the rate check and the high alarm all pass. What is the most likely fault?
MCQ-71A 6 m single-ended type-K thermocouple run on a 6-axis cell shares a cable tray with the servo drive cables. With the drives idle the reading is steady; with the drives energised and moving the signal at the PLC input swings by about 40 mV, and the cell must hold ±1 K. Which change should the engineer make?
MCQ-25A skid-steer robot fuses a 10 Hz lidar with a 200 Hz IMU. Straight-line runs are clean, but in turns the fused attitude is off by about 1.0° at 1.0 rad/s and about 0.5° at 0.5 rad/s. The IMU's static calibration is current, its measured bias is under 0.01 rad/s, and its end-to-end latency is 2 ms. What should the team do?
MCQ-29A warehouse AGV fuses a 100 Hz MEMS IMU with wheel encoders and GNSS. Its 30 s tunnel traverse is continuous, no stop, and GNSS is unavailable throughout; cross-track error at re-acquisition must stay under 1 m. Bench calibration leaves a 0.02 m/s² residual accelerometer bias. Which statement should the review accept?
MCQ-30A differential-drive AGV has a nominal wheel radius of 100 mm and a 600 mm track width. On painted concrete under a 400 kg load, a commanded 50.0 m straight run measures 49.25 m on an external tracker, and a commanded 360° in-place rotation measures 354.6°. Repeating each test three times reproduces both numbers to within 20 mm and 0.2°. Which conclusion and fix do these measurements support?
MCQ-2A mobile base drives at 1.2 m/s. Each lidar scan is stamped at the instant it is taken and reaches the mapper 40 ms later, where the mapper looks up the transform at the processing time instead of at the scan's stamp. Odometry is published at 50 Hz and the lookup interpolates between samples. An engineer proposes to raise the publish rate to 200 Hz. What position error does the transform lookup still introduce?
MCQ-24A pallet-detection camera on a forklift AGV runs at 30 Hz. It reports a nominal 40 ms shutter-to-arrival latency, but the actual delay varies between 30 and 50 ms. The AGV travels at 1.5 m/s. The controller timestamps each detection with the frame's arrival time. At the pallet, the detected edge sits 60 mm ahead of the AGV's believed position, and the forks enter 60 mm high. Which change addresses the cause?
MCQ-91A delta robot picks parts off a conveyor that runs at up to 1.5 m/s. Vision locates each part; the belt encoder is polled by the application over EtherCAT every 20 ms, while the robot's own joint state is timestamped in the drive every 1 ms. Vision latency is measured at 3 ms. A laser tracker measures the placement residual against belt speed: 10 mm at 0.5 m/s, 20 mm at 1.0 m/s, 30 mm at 1.5 m/s. The cell's placement tolerance is ±25 mm. Which conclusion and action fit these measurements?
MCQ-47A ROS 2 control node subscribes to a joint-pose topic published at 100 Hz. The subscription's QoS is reliable, keep-last, depth 10. The subscriber's callback thread is blocked for 300 ms by a garbage-collection pause while the publisher keeps publishing at 100 Hz. When the callback resumes, what does it act on?
MCQ-48A pick-and-place cell misses grasps on fast-moving parts. The camera publishes at 30 Hz, a relay node forwards frames in batches of 6, and the consumer's rate counter reads a steady 30 Hz. The control loop needs content age at the consumer under 50 ms. Which conclusion does the evidence support?
MCQ-49A ROS 2 Humble cell launches 12 nodes from one launch file. The calibration node reads its extrinsics parameter about 40 ms after its process starts; the parameter store finishes loading the YAML anywhere from 200 ms to 2.5 s later, depending on disk and network. The camera is re-mounted between shifts, so the extrinsics change. Over 40 boots, 6 ran with an identity transform and placed parts against an unrotated camera frame. The controller also subscribes to /tf, which the camera driver begins publishing at a variable time. The cell must never cycle with a wrong frame, and a genuinely missing parameter must stop the launch. Which change fixes this?
MCQ-40A 1 kHz position loop on a 6-axis arm has a 1 ms deadline. Under production load it shows a 0.40 ms mean, a 0.95 ms 99.9th percentile, and a 2.30 ms 99.99th percentile; the axis logs a following-error spike about every 10 seconds. Kernel tracing shows every 2.30 ms cycle starts within 50 µs of a cpufreq transition on the core running the control thread. Which change addresses the cause?
MCQ-41A single-core Linux box runs a 1 kHz control thread (SCHED_FIFO 80, 0.3 ms worst-case execution) that shares a mutex with a logging thread (SCHED_FIFO 20). A network handler runs at SCHED_FIFO 50. A trace shows the control thread blocked on the mutex for 3.1 ms, with the network handler running for 2.8 ms of that window; the control loop missed three consecutive deadlines and the axis lurched. Which single change removes the unbounded blocking?
MCQ-87A 6-axis arm runs a 1 kHz control loop (1 ms period) that reads a 200-byte pose estimate published by a 30 Hz perception thread. Both sides take the same mutex; the perception thread holds it for up to 1.4 ms while writing. The loop misses its deadline about 30 times per second and the arm shows a periodic disturbance at 30 Hz. Which conclusion and fix fit these measurements?
MCQ-42A 6-axis arm controller runs a 100 Hz control loop that kicks a hardware watchdog at the end of every cycle. In a field incident the arm held its last commanded velocity for 40 minutes while the pendant showed 'running' and the watchdog never fired. The loop had kept completing cycles, but the CAN driver was returning the same cached joint-position frame every cycle, so the controller recomputed the same command. Which change makes the watchdog fire on this failure?
MCQ-84A 200 Hz arm control loop has a 5 ms cycle budget. Profiling the current stack gives the Python behaviour node p50 0.9 ms and p99.9 34 ms, while the C++ control layer holds p99.9 0.4 ms; during one 34 ms pause the arm held its last command and overshot the handover point by 27 mm. Which approach holds up?
MCQ-76A 1 kHz control loop has a 1 ms budget. Over 10^7 profiled cycles the trajectory buffer's on-demand growth allocates with a median of 0.9 µs, a p99 of 41 µs and a p99.99 of 2800 µs, and every deadline miss lands on a cycle where an allocation ran. Which change addresses the misses?
MCQ-75A candidate logging design for an embedded target is being assessed before commissioning. The control loop runs at 1 kHz with a 1 ms deadline, and the p99.9 iteration cost must stay under 1 ms. Logging at production verbosity must run for hours and survive a power cut; the log is diagnostic, and the team has agreed that dropping a bounded number of records is acceptable as long as the loss is observable. Which design is defensible?
MCQ-51A six-axis arm runs a 1 kHz control loop. Each state record (commanded and measured joint state, mode, limits, safety inputs, fault codes) is 200 bytes. The controller has 8 MB of spare RAM and a 4 GB flash partition for logs. Protective stops are caused by following-error spikes lasting 20–40 ms, and the team needs the 2 s before each stop at control-loop resolution. Which recording design lets them reconstruct the spike and stay within the storage budget?
MCQ-85A trajectory smoother normalises waypoints in place, `seg = waypoints[2:5]; seg /= np.linalg.norm(seg, axis=1, keepdims=True)`, and also zeroes the velocity of `waypoints[0]`. One caller reuses its array for a second execution and gets a different path. The smoother is called from six sites, three of which reuse the array afterwards, and it runs once per control cycle on a block of a few hundred bytes. Which change should the review accept?
MCQ-86A 2D collision checker decides whether a point lies left of a directed edge by the sign of the orient2d determinant (bx−ax)(cy−ay) − (by−ay)(cx−ax), computed in IEEE-754 double. Vertices come from CAD with coordinates up to 2 m, so each product reaches about 16 m². On a 40,000-triangle mesh, 12% of edge-adjacent triangle pairs are reported as overlapping, and swapping the two arguments of the predicate flips the answer for 9% of pairs. What is the correct fix?
MCQ-82A warehouse AMR runs A* on a 40,000-node grid. Over one replanning burst the open list records 260,000 pushes, 190,000 pops, and 125,000 pops whose node already had a lower g-score in the closed set. Peak open-list size is 96,000 entries. The heap is a plain binary heap; the planner pushes a new entry on every improvement and never checks for stale entries on pop. Each pop expands the node and relaxes its neighbours. Replanning p99 is 210 ms and misses the 100 ms control deadline. Which change is most likely to bring p99 under the deadline without changing the search result?
MCQ-81A 250 x 250 grid (62 500 cells) is searched with 4-connected A-star, and obstacles are fixed for the shift. Entering a cell costs 1.0 on concrete and 3.5 on carpet, and carpet covers 80 percent of the map. Start and goal are 11 cells apart in a straight line, with the corridor between them carpet, so a route along it costs 3.5 x 11 = 38.5. h(n) is the straight-line distance in cells. Measured on this map: Dijkstra plans in 130 ms over 61 800 expansions and returns 38.5; this A-star expands 58 400 and also returns 38.5; the same A-star with h multiplied by 1.5 expands 3 100 and returns 41.0 in 9 ms. Planning must return the cheapest route and finish inside 50 ms. Which change should ship?
MCQ-80A mapping team is sizing the occupancy grid for a new robot. The workspace is 30 m × 30 m × 3 m, mapped at 3 cm resolution, and each cell stores a one-byte log-odds value. Measured occupancy is 0.5%. A dense voxel array costs one byte per cell; the octree they are considering costs 40 bytes per leaf node. Which representation should they ship?
MCQ-55A robot cell samples each motion segment at a fixed number of configurations for collision checking. On the fastest segment the tool travels 62 mm between consecutive checks, and the smallest fixture in the cell is 40 mm across. The engineer must justify the resolution setting to a reviewer. Which justification is sound?
MCQ-56A robot cell runs a fixed transfer move (identical start pose, goal pose, and fixture layout every cycle) planned with RRT-Connect. Over 1000 logged cycles the planner returned a median path of 2.10 m in 120 ms, a 99th-percentile path of 2.80 m in 940 ms, and a worst case of 3.40 m in 1.9 s; the motion budget is 1.5 s. About one cycle in 200 tripped the protective field, and every trip occurred on a path longer than 2.6 m that passed within 15 mm of a fixture. The 2.10 m path was validated against the fixture once. Which change removes the trips?
MCQ-57A depalletizing cell plans with MoveIt against a planning scene published by a perception node at 5 Hz; an object not observed for 2 s is removed from the scene. During a pick cycle the arm occludes a 600 mm tote from t=0 to t=3.2 s, and the scene drops it at t=2 s. The next plan is requested at t=3.0 s against a scene stamped 80 ms earlier, and the trajectory drives through the tote's volume. The plan-time scene-age bound is 1 s. Which change removes this failure?
MCQ-58A bin-picking cell uses a 6-axis arm and a parallel-jaw gripper. The grasp detector returns 20 ranked candidates per bin; the planner executes the top-ranked one. Across 200 bin states, the top-ranked grasp was executable in 68 states (34%). Mean retries per part are 2.9, and cycle time p95 is 21 s versus 9 s when the first grasp works. Lowering the detector's score threshold from 0.7 to 0.3 raised top-grasp executability to 41% and cut p95 to 19 s. What should the team do next?
MCQ-5A six-axis palletiser runs a 60-point taught path with the controller's IK set to nearest-to-current-configuration. At rank 34 the solver flips from elbow-up to elbow-down and the elbow sweeps through a structural column that sits in the elbow-down swept volume for ranks 30-40. The column cannot be moved, the tool orientation at each waypoint is fixed by the pallet pattern, and ranks 50-60 are reachable only elbow-down. Which change should the review accept?
MCQ-6A 6-DOF arm tracks a straight Cartesian path at 0.05 m/s, and the path direction near waypoints 20–24 is close to the Jacobian's least-gain direction. Joint 4 is limited to 200 deg/s. Along the commissioned path the smallest singular value of the Jacobian is 0.018 at waypoint 20, 0.012 at waypoint 22, and 0.004 at waypoint 24. The controller inverts the Jacobian with a plain pseudoinverse and clips each joint rate at its limit. Near waypoint 22 the tool lags the path by 9 mm and the operator feels a jerk when the clip releases. Which diagnosis and fix is right?
MCQ-7A 7-axis inspection robot tracks a 900-waypoint Cartesian path for a 6-DOF task. The controller uses a Moore-Penrose pseudoinverse with no null-space term. In a bench run (worked example), joint 3 starts 24° from its upper limit and drifts 0.03° per waypoint toward it at a constant rate, with tool-path error under 0.1 mm. What does the run do?
MCQ-8A 6-axis welding arm runs a resolved-rate (differential IK) controller at 250 Hz. Over a 28 s seam at a commanded tool speed of 0.12 m/s (3.36 m of travel), the linearised solution is accurate to 0.44 µm per cycle, yet the tool finishes 3.1 mm off the taught path. The offset grows monotonically with distance travelled, is unchanged when the same seam is run at half speed or when the servo rate is doubled, and is zero when the arm holds position. The controller commands joint rates and never compares the achieved pose with the commanded pose. Which conclusion does this data support?
MCQ-9A 6-axis arm picks parts off a conveyor whose speed is fixed by the upstream process, so the pick window is 0.6 s wide. The motion planner is configured with Cartesian collision checking and joint position limits only. The controller enforces joint position, velocity (3.0 rad/s per joint, set by the gearbox rating) and acceleration limits on every move, programmed or jogged. A new pick path is collision-free and keeps every joint inside its position limits, but joint 4 must sweep 2.4 rad within the 0.6 s the planner allotted that segment. The planner reports the path feasible. What happens when the path runs?
MCQ-66A 6-axis arm runs a straight-line Cartesian move from pose A to pose B. Both endpoints clear the release threshold: σ_min(J) = 0.28 at A and 0.31 at B, against a limit of 0.02. The move aborts at 47% of the path on every run with a joint 4 velocity limit error, where the commanded joint 4 rate is 3× the limit. What is the correct next step?
MCQ-54A 6-axis arm follows a 200 mm straight-line path. The controller time-parameterises the path in Cartesian space, limiting the jerk of the path parameter to 5000 mm/s³. Joint 3's rated jerk is 5000 °/s³. Along this path the Jacobian maps tool motion to joint 3 at 4 °/mm at the start and 6 °/mm at the end. After each stop the tool rings at 22 Hz with 0.8 mm amplitude for 300 ms, and the camera images the part 150 ms after the motion-complete flag. Which change removes the ring?
MCQ-77A Cortex-M0 board with no FPU runs a 1 kHz PI speed loop. The integral term is a Q15 accumulator: a signed 16-bit value scaled by 1/32768, range -32768 to 32767, and each tick it adds 1024 counts while the speed error is saturated at its maximum. A stall holds the error saturated for 100 ms. What should the review conclude about the integral term during the stall?
MCQ-23A joint controller tuned at 1 kHz uses Kp = 40, Ki = 200, Kd = 0.5, with the integral accumulated as I += Ki·e and the derivative as Kd·(e − e_prev), each once per cycle. The drive is replaced by one that runs the same loop at 250 Hz, keeping the gains and the per-cycle accumulation. Under a constant gravity load the joint now settles with 0.02 rad of steady-state error, up from 0.005 rad, and takes about four times as long to null it. What explains the change, and what should be done?
MCQ-22A robot arm's joint-2 velocity loop uses a proportional controller with gain Kp. Open-loop step tests give the process gain K (rad/s per unit command) at three joint angles: 0.40 at 0°, 0.90 at 45°, and 1.80 at 90°. The loop is well damped when the open-loop crossover Kp·K is 8 rad/s and oscillates once crossover passes 12 rad/s. A single tune set at 45° oscillates at 90° and is sluggish at 0°. Which Kp schedule holds crossover at 8 rad/s across the range?
MCQ-21A jacketed batch reactor runs a cascade: an inner PI loop holds jacket temperature by modulating steam flow, and an outer PI loop holds reactor temperature by trimming the jacket setpoint. Neither controller uses derivative. With the outer loop in manual, the jacket loop holds setpoint within ±0.2 °C with no oscillation and a closed-loop time constant of 20 s. The outer loop is then tuned to a closed-loop time constant of 30 s. On the heat-up test the reactor temperature oscillates with a period of about 2 min and ±1.8 °C amplitude; the jacket setpoint stays within ±3 °C of nominal and never approaches its ±10 °C saturation limits. Which action should the review require before the next heat-up test?
MCQ-16An oven loop holds 200 °C. From a cold start the heater sits at its 100 % output limit for about six minutes, then the temperature overshoots to 231 °C before settling, scrapping the parts already inside. Which change addresses the cause of the overshoot?
MCQ-19A cascade flow/temperature loop runs on a PLC with a 100 ms scan; the inner flow loop has a 2 s integral time and the outer temperature loop 60 s. The operator has held the flow valve in manual at 41% for 20 minutes because that value keeps the product in spec, and the process has drifted so that with the integral term at zero the controller would command 62%. The operator enters a new setpoint and transfers to auto. Commissioning requires the valve to stay at 41% until the controller acts on the new setpoint, and the new setpoint to be acted on from the first scan. Which transfer design meets both requirements?
MCQ-18A steam-header temperature loop has a measured dominant time constant τ = 8 s and a 12 s transport lag in the process. The PID was tuned to a 30 s settling target; it cycles with a 45 s period and trips the high-temperature limit. The sensor, actuator, and network are healthy. What should the team do next?
MCQ-17A hydraulic servo axis runs a position loop at 1 kHz with a 12-bit encoder (4096 counts/rev). Standing still, the encoder reading jitters ±2 counts sample to sample. The controller uses Kp = 0.25 %/count and derivative-on-error with Kd = 4 %/(count/sample), unfiltered. The valve command ripples ±8% at 90 Hz while standing still, and on a setpoint step it saturates for one 1 ms sample before settling with 5% overshoot in 40 ms. The spool logs 2.1 million reversals in a week against a 5 million rated life. Which change addresses both symptoms?
MCQ-15A six-axis arm's axis 3 is rebuilt and re-enabled for the first time with the drive torque limited to 1.2 N·m (10% of its 12 N·m peak). The engineer commands a +0.10 rad step. The axis physically rotates in the + direction and reaches the + hard stop in 180 ms; over the same interval the controller's reported position for axis 3 falls monotonically from 0 to −0.42 rad, and the drive faults on following error as the axis reaches the stop. Which conclusion is correct?
MCQ-13You are reviewing a controls change for a press-fit station. The motor encoders hold 0.02 mm at every load, but a dial gauge at the tool reads 0.02 mm at 0 N, 0.21 mm at 30 N and 0.41 mm at 60 N of insertion force. Insertion force varies from 20 N to 60 N across the part mix, the seat tolerance is ±0.15 mm at the tool, the axis already runs at the gain limit that keeps it from oscillating, and the seat is approached from whichever side the preceding move leaves. Which statement should the review accept?
MCQ-59A 6-axis arm presses a part into a fixture. Its position loop has an effective stiffness of 50 kN/m. Contact stiffness measured on the three surfaces in the cell is 2 kN/m (foam), 40 kN/m (plastic) and 900 kN/m (steel). On the steel fixture the loop chatters at about 60 Hz and has scrapped 400 parts; foam and plastic run clean. Which conclusion is correct?
MCQ-92A 6-axis robot's tool axis has a measured 14 Hz resonance with damping ratio 0.02. The application needs 9 Hz closed-loop bandwidth to make cycle time. A 14 Hz notch removes the ringing, but during 9 Hz moves the axis lags and cycle time rises 18%. The team must ship the cell. What should they do next?
MCQ-79A 6-axis arm's controller runs its production firmware at 1 kHz on the real EtherCAT bus, with the arm dynamics simulated. The rig feeds the controller an ideal encoder position at every 1 ms tick, and the simulated dynamics were validated against the real arm to within 3% on the test trajectory. All 240 acceptance tests pass on the rig. On the real cell the arm trips a following-error fault during its first high-speed move: the production encoder reports position every 250 µs with ±40 µs of update jitter, and the drive trips when following error exceeds 0.05° for 2 ms. What is the sound next step?
MCQ-20A conveyor-tracking robot must place parts while the conveyor runs anywhere from 0.1 to 0.5 m/s, against a ±1 mm placement tolerance. Measured tracking lag is 2 mm at 0.1 m/s, 6 mm at 0.3 m/s and 10 mm at 0.5 m/s. Raising the position loop's proportional gain from 12 to 40 cut the lag at 0.5 m/s to 2 mm, but the axis then oscillated at 14 Hz on every conveyor jog and placement got worse than with the original lag. Which approach should the team take?
MCQ-31A palletising cell runs an AGV fitted with a safety laser scanner. The datasheet gives the protective field range as 4.0 m against a 90% reflective target and 3.1 m against the 1.8% matt black test body; at 18% reflectivity it is 3.4 m. The AGV's measured worst-case stopping distance, including scanner response time and brake wear, is 2.6 m. The cell handles matt black plastic pallets (1.8%) alongside bare aluminium ones. What is the largest protective field radius the scanner can be trusted to enforce in this cell?
MCQ-38A tugger AGV carries a 2D safety laser scanner on its front face with a 1.0 m protective field radius, scan plane 200 mm above the floor. It tows a cart whose load overhangs 1.2 m ahead of the scanner, riding above the scan plane. A person stands 2.0 m directly ahead of the scanner. As the tugger advances, the scanner reports the path clear. Which action is correct?
MCQ-39An AMR fuses a front 2D safety lidar and a depth camera. The arbitration rule is: if the two ranges differ by more than 0.3 m, command a stop and use the smaller range. Baseline disagreement is 2/hour. On day 5 it is 31/hour and on day 11 it is 140/hour; the camera lens is found fogged on day 11. Over that window the robot logged 1,240 stops, and in 1,180 of them the camera reported the smaller range. What does this data show?
MCQ-35A 2D-lidar SLAM stack on a warehouse AMR is being commissioned. Bench runs measure 1.5% odometry drift over distance travelled, inside the 2% spec. On a 140 m round trip down an aisle of identical pallet bays, the map shows the aisle twice, offset 2.1 m laterally, and the planner routes through a rack. The run log records 0 accepted loop closures and 38 candidate matches rejected by the place-recognition descriptor. Which action is right?
MCQ-36A 300 kg AMR approaches a pallet rack in a live cell. Its filter reports a 1σ lateral position uncertainty of 0.05 m at the docking approach, which clears the site's speed gate (<0.10 m → 1.5 m/s; 0.10–0.35 m → 0.6 m/s; >0.35 m → re-localise) for 1.5 m/s. A 200-run comparison against a laser tracker puts the actual 1σ error at 0.30 m. What should the team do before the cell runs at full speed?
MCQ-37An AMR localises against a 2D lidar map, and its scan matcher needs at least 300 inliers to lock reliably. Over three weeks one aisle is re-racked, moving about 40% of the structure the matcher sees. Match rate stays above 0.9 and the robots keep completing cycles, but the median inlier count against the map falls 780 → 460 → 210 and the 5th-percentile count falls 410 → 60. On day 22 a robot stops mid-aisle with a localisation failure and blocks the shift. What should the team have done?
MCQ-70A 24 V LiFePO4 pack on a warehouse AMR estimates state of charge from terminal voltage alone. Bench discharge tests record: at rest the pack reads 26.1 V and reports 38%, matching the 38% of rated energy it actually holds; under a 20 A climb it reads 24.6 V and reports 30% while it actually holds 37%. The dispatcher clears a robot for a task needing 35% off the rested dock reading of 38%; mid-route the estimate falls to 30%, the robot stops, and it blocks the aisle for 40 minutes. What is the correct diagnosis and fix?
MCQ-43A vertical axis carries a 60 kg fixture on a power-off brake. On the machine, the E-stop disables the servo at 15 ms and the brake engages at 135 ms; the fixture drops 40 mm in the gap and strikes the fixture plate. The hazard analysis requires the axis to hold position through the stop. Which change is defensible?
MCQ-44A robot cell's speed limit is computed by the same motion controller that generates the trajectory. The risk assessment assigns the speed-limiting function PL d, Category 3, and a units-conversion defect once raised the commanded speed and the limit together: the cell ran at 3.2× the intended speed with no trip. Which design satisfies the assessment?
MCQ-45A cell runs a 6-axis arm the manufacturer rates for collaborative operation. The integrator's risk-assessment file records the force and pressure check against the 900 mm² bare flange, but the end effector now carries a 12 mm probe tip. The line is due to restart. Which action should the review take before releasing the file?
MCQ-46A collaborative cell is being commissioned. The layout is fixed: the closest an operator's hand can reach is 780 mm from the robot, and that distance cannot be increased. The robot runs at 1.2 m/s tool speed. The integrator takes the human approach speed as 1.6 m/s and the end-to-end system reaction time as 300 ms. With the 12 kg payload fitted, the measured stopping distance is 310 mm; sensor and position uncertainty is 120 mm. What must be done before the cell is allowed to run?
MCQ-72A cell washes a 6-axis arm after every shift with 60 °C caustic at 3 bar from a nozzle held 300 mm from the tool. The arm is IP65, the tool connector IP54, and the cable gland IP67. After three weeks the connector pins are corroded; the arm and gland show no damage. Which change addresses the cause?
MCQ-74A palletizing arm streams motion frames from a fleet server over Wi-Fi at 20 Hz. Each frame carries a 32-bit sequence number, a send timestamp, and a `state` field; `state = RUN` energises the drives and `state = HOLD` drops them. The local controller rejects any frame whose sequence number is not greater than the last accepted one, or whose timestamp is more than 150 ms behind its own clock (NTP-synced to the server's within 5 ms). It also latches a protective stop when the cell gate opens, and that stop blocks motion while it is set. At t = 0 the link dropped for 40 s. At t = 12 s an operator opened the gate to clear a jam: the drives dropped, the stop latched, and the gate was still open when the link recovered at t = 40 s. The server, which has no view of the gate, buffered 40 s of motion frames and replayed all 800 in order on reconnect, then resumed its live 20 Hz stream. Log: the replayed frames were dropped as stale except the last, 20 ms old and repeating a pose the head already held. The first live frame, 40 ms after the link recovered, carried a new sequence number and `state = RUN`; the latched stop cleared, the drives energised, and the arm resumed its cycle with the gate open. Which change prevents that?
MCQ-73You run OTA for a fleet of 60 AGVs in two hardware revisions: 22 on rev A, 38 on rev B. A canary of 3 rev-B units takes the new image and each passes the 90-second post-boot health check, so you widen to all 38 rev-B units. Within the hour, 11 of them fault with a motor-controller watchdog trip: every one between 35 and 50 minutes after boot, on its first sustained drive above 0.6 m/s. The canary units had run the standard test route, which never exceeds 0.4 m/s. Rev-A units are untouched. What do you do next?
MCQ-78A 6-DOF arm's motion stack runs 400 unit tests against a mocked controller interface; all pass in CI. On the real arm the controller rejects every trajectory with 'position out of range': the planner emits degrees, the controller expects radians, and the mock accepts any value it is handed. The integration simulator shares that same interface stub. A previous release oscillated at 8 Hz under a 20 N contact load and no test caught it. Which change to the test strategy addresses both the escaped defect and the contact risk?
MCQ-60A 6-axis wrist force/torque sensor carries a 1.4 kg tool. With the cell's 10 N contact threshold, the robot reports contact in mid-air on about one approach in four. With no contact, the raw reading swings from +14 N with the tool down to −13.6 N with the tool up, and the false trips grow over the first hour as the cell warms up. The process needs to detect contacts as light as 10 N, and the cell has to run at full speed next week. Which approach holds up?
MCQ-61A two-cavity mould feeds a pick-and-place cell. The parallel two-finger gripper opens to a maximum of 24.40 mm, and its close position is set in the robot program. Tuned on 200 cavity-1 parts, all 24.10 mm ±0.05 mm, it picks them at 99%. Cavity-2 parts run 24.45 mm ±0.15 mm; on those the fingers meet the part as the gripper opens and the pick rate is 41%. Pooled across both cavities the cell reports 78%. The integrator proposes re-teaching the grasp pose on cavity-2 parts and shipping. What should the review decide?
MCQ-62A pick-and-place cell runs an 11.5 s cycle against a 9.0 s takt. Measured intervals: approach to the pick station 2.6 s; vision exposure and inference 2.6 s, triggered only after the approach settles; gripper close 1.2 s; transfer to the place station 3.1 s; place and return 2.0 s. The camera is fixed above the pick station and the part is in frame for the whole approach, and the vision result sets the grip correction, so it must complete before the gripper closes. Which change brings the cycle under takt?
MCQ-64A six-axis cell runs a weekly reference move at fixed speed and payload, logging peak current and lost motion per axis. At commissioning, axis 3 read 4.2 A / 0.04° and axis 5 read 5.4 A / 0.05°. At week 6, axis 3 reads 5.6 A / 0.21° and axis 5 reads 5.5 A / 0.05°. The maintenance plan alarms on absolute peak current at 6.0 A; the fleet mean is 5.4 A. What should the team do?
MCQ-65A six-axis arm with a 12 kg end effector runs a pick-and-place cycle in which the same corner is traversed at 30% to 100% speed depending on the part mix. The taught path clears a welded steel fixture by 8 mm, and the fixture cannot be moved. The program was verified at 20% speed on the bench, and the offline simulation used for reach and cycle-time checks is kinematic and has never been compared against measured following error. This cell's arm is mounted on a different pedestal than the sister cell already in production. Which commissioning approach should the team adopt?
MCQ-67A cell's task sequencer is a state machine with 11 states. The current state is kept only in RAM and there is no recovery transition. After an e-stop the controller restarts, the machine enters its initial state startup, and its first transition drives to the home pose. In 4 of the 11 states the gripper holds a part, and restarting from those states strikes the fixture on the way home. The team adds one guard: the startup → home transition is refused while the gripper sensor reports a part. Stops are then injected at all 11 states, the fault is cleared, and the controller is restarted. What is the outcome?
MCQ-68A bilateral teleoperation cell runs stably at 20 ms round trip, but when the operator's traffic is routed over a VPN at 95 ms the force loop breaks into a 4 Hz oscillation. Which change holds up?
MCQ-69Two AMRs with 1.1 m footprints approach a 1.6 m wide, 12 m long aisle that is the only route to the pick face, one from each end. Each robot's local planner sees the other, marks the aisle blocked, and stops. Both keep reporting a valid local plan and no fault. After 26 minutes an operator backs one robot out. In the same layout, 2.6 m wide aisles never show this. Which change should the controls review accept to prevent the 1.6 m aisle case?
MCQ-50Two identical cells run release 4.2.1 and report the same code commit. Cell A cycles at 8.4 s; cell B at 11.9 s. Their on-disk configuration files are byte-identical, their commissioning records match the released baseline, their commanded tool paths in rviz are identical, and swapping cell B's axis with a known-good unit leaves the cycle time unchanged. The release manifest records only the code commit. Which change to the release process would have caught this at the moment it was introduced?
MCQ-52A bin-picking policy is trained in simulation with the part's friction coefficient fixed at 0.8. In simulation it scores 98% whether the simulated friction is set to 0.35 or to 0.9. On the real cell it succeeds on 71% of 500 trials. Logging each part's measured friction: 0.35–0.45 → 21% success, 0.55–0.65 → 74%, 0.85–0.95 → 96%. Production needs first-attempt success above 90% across the full 0.35–0.95 range, and the cell runs at full rate in one week. What is the right next move?
MCQ-53A manipulation policy trained with domain randomisation over lighting, texture, camera pose and joint friction reaches 90% success in simulation and 62% on hardware. Measured joint friction across the fleet sits inside the randomised range, and the hardware logs show a 30 ms actuation delay that the simulator's actuator model, a pure gain, does not represent. The team proposes tripling the randomisation ranges. What should the review require?
MCQ-93A robotic placement cell must place a 2.0 kg part within ±0.5 mm at a 12 s cycle time. The integrator proposes acceptance: 50 consecutive cycles at 80% speed using parts from one supplier's lot, all within tolerance, no fault injection. Production will run three shifts with parts from three suppliers whose dimensions span the full tolerance band. Which change should the review require before sign-off?
MCQ-95A pick-and-place cell runs three cameras feeding a pose estimator. Its validated envelope is: 3 cameras → full position and orientation at 100% rate (400 parts/h); 2 cameras → position only, orientation resolvable for variant A but not variant B, permitted rate 60%; 1 camera → stop. At 14:00 one camera fails; the controller keeps reporting 'normal operation' and the line continues at 400 parts/h. The shift ends at 22:00, 1,500 variant-A parts must ship, and variant B can be deferred. Variant B is 30% of the normal mix. What should the engineer do?
MCQ-96A cell's maintenance log for the past year shows two components dominating downtime. The camera mount failed twice; each replacement took 3.5 h, most of it re-aligning the mount and re-running extrinsic calibration. The gripper failed six times; each swap took 10 minutes. A keyed mount with stored extrinsics is available for the camera and swaps in 12 minutes with no recalibration. The cell is scheduled 4,000 h/yr and carries a 0.1% downtime budget. Which change removes the most downtime per year?
MCQ-97A six-axis arm is being specified for a deburring cell. The datasheet lists a 12 kg rated payload, ±0.05 mm repeatability and 2 m/s tool speed. The tool and part together weigh 10 kg, the tool's centre of mass sits 180 mm from the flange, the part tolerance is ±0.1 mm across a 900 mm working envelope, and the mounting plate and 1.2 s cycle budget rule out the next frame size up. Which approach should the team adopt?
MCQ-98A two-variant cell changes over by swapping gripper fingers, loading a variant program, and changing a fixture plate. The fixture plate is keyed and cannot be fitted to the wrong variant. Over six weeks the cell ran 180 changeovers; 9 produced scrap, 3 to 8 parts each, always the first parts after a changeover. In all 9 events the variant program had loaded correctly and the operator's changeover checklist was signed off. The cell runs unattended for the first 20 minutes after each changeover while the operator moves to the next cell. Which change to the changeover should be made first?
MCQ-99A palletizing cell collided with a safety fence at 03:14 and stopped. The postmortem, written at 09:00 the same morning from the planner's 10 Hz decision log, concludes 'servo fault': the log shows a smooth commanded path right up to the stop. The controller's 1 kHz motion buffer is a 30-minute ring. The report is queued for acceptance. Which action is correct?
MCQ-100A shop is quoting a robot cell to load a CNC lathe. Installed cost (robot, gripper, guarding, integration) is $120,000. The lathe runs manually today with one operator at $85,000 a year fully burdened. The vendor's quote assumes parts arrive in two presentations at a 4% exception rate, which leaves 0.1 FTE of operator attention and, with $12,000 a year of maintenance, gives a 22-month payback. A two-week observation of incoming parts found eleven presentations and a 19% exception rate, which leaves 1.0 FTE of operator attention. The shop requires payback inside 24 months. What does the observed data support?