Tricky Robotics & Automation interview questions

75 questions with answers (RAA-426 to RAA-500). Try each one aloud before reading the answer.

Robot Fundamentals & Classification

RAA-426 · Why do robot arms typically need 6 DOF? Isn't 3 enough to reach any point?

Trap: Reaching a point is not the same as reaching a pose. Three DOF can position the tool, but three more are needed for orientation, so 6 DOF is the minimum to achieve an arbitrary pose in 3D.

RAA-427 · Is more DOF always better for a robot?

Trap: Assuming flexibility is free. Redundancy helps avoid obstacles and singularities, but adds cost, weight, compliance, control complexity, extra singularity configurations and maintenance. Use the fewest DOF the task needs.

RAA-428 · A SCARA has 4 joints, so does it have 4 DOF in any task?

Trap: Counting joints as task DOF. SCARA provides X, Y, Z and yaw only; it cannot tilt the tool, so tasks needing roll or pitch are impossible regardless of programming.

RAA-429 · Does a robot with 0.02 mm repeatability place parts within 0.02 mm?

Trap: Confusing robot repeatability with process accuracy. Gripper slip, part tolerance, fixture variation, calibration and thermal drift add up. The overall placement error is usually much larger than the robot's repeatability.

RAA-430 · Does a higher rated payload always mean the robot can carry that weight anywhere in its workspace?

Trap: Ignoring centre-of-gravity offset and inertia. Rated payload applies at a specified distance from the flange; larger offsets or inertia reduce allowable mass per the payload diagram.

RAA-431 · Is a parallel robot always more accurate than a serial robot?

Trap: Equating stiffness with accuracy. Parallel robots are stiffer and errors don't accumulate along a chain, but their kinematics are complex, the workspace small, and accuracy degrades sharply near their singularities.

RAA-432 · Is a robot with a prismatic joint a Cartesian robot?

Trap: Classifying by one joint type. The configuration depends on all major axes: a cylindrical (RPP) and spherical (RRP) robot also have prismatic joints. Only three orthogonal prismatic axes (PPP) make a Cartesian robot.

Forward & Inverse Kinematics

RAA-433 · Can inverse kinematics have no solution?

Trap: Assuming every commanded pose is achievable. IK has no solution if the target is outside the workspace, the orientation is impossible at that position or it requires violating joint limits.

RAA-434 · Does the IK solution with the smallest joint angles always give the best motion?

Trap: Minimising joint values instead of motion. The best choice is usually closest to the current configuration and away from limits and singularities; a "small angle" solution may require a huge flip of the arm.

RAA-435 · At a singularity, can the robot not move at all?

Trap: Thinking a singularity freezes the robot. It loses only one or more Cartesian directions instantaneously; it can still move in other directions and in joint space.

RAA-436 · Are singularities only at workspace boundaries?

Trap: Ignoring internal singularities. Wrist and shoulder singularities occur inside the workspace, such as when axes 4 and 6 align, and are the most common cause of Cartesian motion faults.

RAA-437 · If two DH tables describe the same robot, must their parameters be identical?

Trap: Assuming DH is unique. Frame choices (origins, direction of z or x, standard vs modified convention) are partly arbitrary, so different valid tables exist, but they give the same tool pose.

RAA-438 · Do Euler angles give a unique description of every orientation?

Trap: Ignoring representation singularities. At gimbal lock, infinitely many Euler angle sets describe the same orientation, and angles wrap at ±180°. Quaternions or rotation matrices avoid this.

RAA-439 · Does inverting the Jacobian always give the joint velocities for a desired tool velocity?

Trap: Assuming J is square and invertible. J is non-square for redundant or under-actuated arms and singular at singularities; the pseudoinverse or damped least squares must be used.

Robot Dynamics

RAA-440 · Is the torque needed to hold an arm still zero because it isn't moving?

Trap: Forgetting gravity. Static holding requires G(q), which is maximal with the arm stretched horizontally. Only brakes, counterbalances or a vertical pose reduce it.

RAA-441 · Does doubling a joint's speed double the Coriolis and centrifugal torques?

Trap: Assuming linear dependence. These terms depend on products of velocities, so doubling all joint speeds quadruples them.

RAA-442 · Does a higher gear ratio always make a robot better?

Trap: Ignoring the downsides. Higher ratios increase torque and decouple joints but reduce top speed, add friction and reduce backdrivability, making force control and safe contact harder.

RAA-443 · Is the inertia seen by the motor equal to the load inertia?

Trap: Forgetting the gear ratio. The load is reflected as J_load/N², so with high reduction the motor's own rotor inertia often dominates.

RAA-444 · If a dynamic model is accurate for an empty arm, does it stay accurate with a gripper attached?

Trap: Ignoring payload parameters. Tool mass, centre of gravity and inertia change gravity and inertial terms; they must be entered or identified, or tracking and collision detection degrade.

RAA-445 · Does a motor that meets the peak torque requirement always suffice?

Trap: Checking only peak torque. The motor must also meet RMS (continuous) torque for heating, speed limits and inertia ratio; a cycle can be within peak but overheat the motor.

Trajectory Planning

RAA-446 · Is PTP motion the shortest path between two points?

Trap: Equating joint interpolation with straight lines. PTP is typically fastest in time, but the TCP traces a curve, which may collide with nearby obstacles.

RAA-447 · Is a straight-line Cartesian move always safer than a joint move?

Trap: Assuming straight is safe. A LIN move can pass near singularities, producing very high joint speeds, or demand large wrist reorientation. Safety depends on the path relative to obstacles and robot configuration.

RAA-448 · Does setting maximum acceleration always give the fastest cycle?

Trap: Ignoring vibration and settling. Very high acceleration excites resonances, adding settling time and wear; jerk-limited profiles often finish faster in practice.

RAA-449 · Does a cubic polynomial trajectory produce smooth motion at its endpoints?

Trap: Assuming position and velocity continuity is enough. With zero end velocities, a cubic has non-zero acceleration jumps at start and end, giving infinite jerk; quintics fix this.

RAA-450 · Is halving the programmed speed of a move going to halve the forces on the robot?

Trap: Assuming linear scaling. Inertial and centrifugal forces scale with velocity squared for a fixed path, so halving speed cuts them to about a quarter, while gravity loads are unchanged.

RAA-451 · Does a robot always pass exactly through a programmed via point?

Trap: Ignoring blending. With a zone or approximation setting the robot cuts the corner and does not reach the point; a fine (stop) point is needed for exact positioning.

Robot Control

RAA-452 · Does increasing proportional gain always improve tracking?

Trap: Thinking more gain is always better. Higher Kₚ reduces error until phase lag, delays and resonances cause overshoot, oscillation and instability; noise and actuator saturation also limit it.

RAA-453 · Does integral action always help?

Trap: Treating I as free. It removes steady-state error but adds phase lag, can cause overshoot and windup during saturation, and may cause limit cycles with friction or backlash.

RAA-454 · Can a pure position-controlled robot press with a precise force on a rigid surface?

Trap: Assuming position control controls force. On stiff contact, tiny position errors cause huge force changes. Force, impedance or admittance control with sensing is needed.

RAA-455 · Is computed torque control robust because it uses the full model?

Trap: Assuming model-based means robust. Its performance depends on model accuracy; payload or friction errors leave residual nonlinear coupling, so it is paired with robust feedback gains.

RAA-456 · Does a stable closed-loop system mean the robot won't oscillate?

Trap: Confusing stability with good damping. A stable system can still have low phase margin, giving ringing and overshoot; damping and margins matter, not just stability.

RAA-457 · Does derivative action on error always reduce overshoot?

Trap: Ignoring derivative kick and noise. Derivative on error spikes on setpoint steps and amplifies noise; derivative on measurement with filtering is preferred.

Sensors & Perception

RAA-458 · Does a higher-resolution encoder give a more accurate robot?

Trap: Confusing resolution with accuracy. Resolution is the smallest detectable step; accuracy also depends on encoder errors, gearbox backlash, link deflection and calibration.

RAA-459 · Can an accelerometer measure tilt during fast motion?

Trap: Assuming it measures gravity only. It measures specific force, gravity plus motion acceleration, so tilt estimates are corrupted during acceleration; gyro fusion is needed.

RAA-460 · Does an inductive proximity sensor detect all metals at its rated range?

Trap: Assuming one sensing distance. Rated range refers to mild steel; aluminium, brass and copper have correction factors that substantially reduce range, unless factor-1 sensors are used.

RAA-461 · Does adding more sensors always improve state estimation?

Trap: Assuming more data equals better estimates. Poorly calibrated, unsynchronised or wrongly modelled sensors can corrupt the fused estimate; noise covariances and outlier rejection must be correct.

RAA-462 · Does a Kalman filter always give the optimal estimate?

Trap: Ignoring assumptions. It is optimal only for linear systems with Gaussian, correctly modelled noise. EKFs on nonlinear systems can diverge with poor linearisation or wrong covariances.

RAA-463 · Does a LiDAR see everything in front of it?

Trap: Assuming full visibility. It misses glass, mirrors and very dark surfaces, sees only its scan plane (2D), suffers occlusion and has gaps between beams at long range.

Computer Vision for Robotics

RAA-464 · Does a higher-megapixel camera always give better measurement accuracy?

Trap: Ignoring optics and lighting. Lens resolution, distortion, focus, depth of field, lighting contrast and calibration often limit accuracy before pixel count does.

RAA-465 · Can a single calibrated camera measure the 3D position of an unknown object?

Trap: Forgetting scale ambiguity. A single view gives only a ray per pixel. You need known object size, a known plane (like a conveyor surface), stereo, motion or a depth sensor.

RAA-466 · Does a deep learning detector trained to 99 % accuracy guarantee 99 % in the factory?

Trap: Assuming test accuracy transfers. Domain shift from lighting, new part variants, camera changes and rare defects reduces field performance; continuous monitoring and retraining are needed.

RAA-467 · Is camera calibration permanent once done?

Trap: Treating calibration as one-off. Vibration, refocusing, temperature, lens adjustments and mount bumps change intrinsics and extrinsics, so periodic verification is required.

RAA-468 · Does a stereo camera work well on a plain white wall?

Trap: Assuming stereo always gives depth. Passive stereo needs texture for matching; uniform surfaces give no disparity. Active stereo projects a pattern to fix this.

RAA-469 · Is a pixel-accurate measurement limited to 1 pixel resolution?

Trap: Thinking accuracy can't beat pixel size. Sub-pixel edge and centroid algorithms interpolate intensity to reach a fraction of a pixel, given good contrast and calibration.

Path Planning & Navigation

RAA-470 · Does A* always find the shortest path?

Trap: Ignoring heuristic conditions. A* is optimal only with an admissible heuristic; an overestimating heuristic gives faster but possibly suboptimal paths. Also, "shortest" is only on the given graph.

RAA-471 · Does RRT give the optimal path if you run it long enough?

Trap: Confusing RRT with RRT*. Plain RRT converges to a suboptimal path; RRT* with rewiring is asymptotically optimal.

RAA-472 · If the planner finds no path, does no path exist?

Trap: Ignoring completeness limits. Sampling planners may not find narrow passages in limited time, and grid planners may miss gaps smaller than the resolution or inflation.

RAA-473 · Is the shortest path always the best path for a mobile robot?

Trap: Optimising only length. Paths hugging obstacles are risky and slow; costmaps trade length for clearance, smoothness, time, and comfort near people.

RAA-474 · Can a differential-drive robot follow any path that a holonomic robot can?

Trap: Ignoring non-holonomic constraints. It can reach any pose, but cannot move sideways instantly; paths with lateral motion must be replaced by turn-in-place or curved manoeuvres.

RAA-475 · Does planning in Cartesian space make arm collision avoidance easy?

Trap: Forgetting the whole arm. Clearance of the tool does not guarantee clearance of elbow and links; collision checks must cover the full robot geometry in configuration space.

Mobile Robots & SLAM

RAA-476 · Is odometry useless because it drifts?

Trap: Rejecting a drifting sensor. Odometry is smooth and accurate over short distances; fused with absolute corrections from LiDAR, vision or GPS, it is essential for localisation.

RAA-477 · Once a SLAM map is built, is SLAM no longer needed?

Trap: Assuming environments are static. Moved racks, pallets and doors change the map; robots switch to localisation on a fixed map, but long-term operation needs map updates or lifelong SLAM.

RAA-478 · Does GPS solve localisation for outdoor robots?

Trap: Trusting GPS blindly. Standard GPS accuracy is metres, with multipath and dropouts near buildings and trees; RTK, IMU, wheel odometry and perception are needed for lane-level accuracy.

RAA-479 · Does more particles always make a particle filter better?

Trap: Ignoring cost and model errors. More particles raise computation linearly, and a poor motion or sensor model still gives wrong estimates; adaptive sampling balances accuracy and cost.

RAA-480 · Can monocular visual SLAM recover true metric scale?

Trap: Assuming one camera gives metric maps. Monocular SLAM is scale-ambiguous; scale comes from IMU fusion, stereo, known objects or wheel odometry.

RAA-481 · Do mecanum-wheeled robots have better odometry than differential drives?

Trap: Assuming omnidirectional means accurate. Roller slip on mecanum wheels makes wheel odometry poor, especially sideways; they often need extra sensors for reliable localisation.

ROS & ROS 2

RAA-482 · Is ROS an operating system?

Trap: Taking the name literally. ROS is middleware plus tools and libraries running on an operating system, usually Linux; it does not manage hardware or schedule processes itself.

RAA-483 · Is ROS 2 real-time out of the box?

Trap: Assuming DDS means hard real time. ROS 2 is designed to enable real time, but it requires a real-time OS kernel, careful executor and memory design, and suitable DDS configuration.

RAA-484 · Does publishing on a topic guarantee the subscriber receives every message?

Trap: Assuming reliable delivery. With best-effort QoS, limited history depth or a slow subscriber, messages are dropped; topics are not a guaranteed queue.

RAA-485 · Should you use a service for a 30-second robot motion?

Trap: Using the wrong interface for long tasks. Services block without feedback or cancellation; actions are designed for long-running goals with feedback and preemption.

RAA-486 · Can ROS 1 and ROS 2 nodes talk to each other directly?

Trap: Assuming compatibility. They use different middleware; communication needs a bridge such as ros1_bridge that translates messages between them.

RAA-487 · If a robot works in Gazebo, will it work the same on hardware?

Trap: Trusting simulation fidelity. Friction, latency, sensor noise, actuator limits and calibration differ; the "sim-to-real gap" requires hardware testing and tuning.

Industrial Automation (PLC, SCADA, Cells)

RAA-488 · Is automation always cheaper than manual labour?

Trap: Ignoring volume and variability. High-mix, low-volume production or frequent product changes may not justify capital, integration, changeover and maintenance costs; manual or hybrid cells can be cheaper.

RAA-489 · Does a PLC execute ladder rungs simultaneously like real relays?

Trap: Treating ladder as parallel hardware. A PLC scans rungs sequentially, top to bottom, so rung order matters and outputs update only at the end of the scan.

RAA-490 · If the same output coil is used in two rungs, will both conditions control it?

Trap: Expecting OR behaviour. The last rung scanned overwrites the earlier one (double-coil syndrome); combine conditions in one rung or use set/reset deliberately.

RAA-491 · Is a PLC input instantly seen by the program when a sensor changes?

Trap: Ignoring scan time and filtering. Inputs are read at the start of the scan and pass through input filters; very short pulses can be missed unless high-speed inputs or interrupts are used.

RAA-492 · Is SCADA used to perform the real-time control loop?

Trap: Confusing supervision with control. SCADA monitors, logs and issues setpoints; time-critical control runs in PLCs, RTUs or controllers, which keep operating if SCADA fails.

RAA-493 · Does automating a bad process make it good?

Trap: Automating waste. Automating an unstable or poorly designed process just makes defects faster; simplify and stabilise the process first.

RAA-494 · Is a 0 mA reading from a 4–20 mA transmitter a measurement of zero?

Trap: Forgetting the live zero. 4 mA represents the lowest value; 0 mA or values below about 3.6 mA indicate a fault such as a broken wire or failed transmitter.

Collaborative Robots & Safety

RAA-495 · Why do industrial robots have cages but cobots often don't?

Trap: Thinking cobots are inherently safe. Cobots limit force and speed and detect contact, so after a risk assessment they can work beside people. The application, not the robot type, decides whether guarding can be removed.

RAA-496 · Is a cobot always slower than an industrial robot in a cell?

Trap: Assuming cobots are permanently slow. Cobots run reduced speed only when people are nearby; with scanners, many can run at full speed when the area is clear, though their top speed and payload are still lower.

RAA-497 · Does pressing the E-stop make a robot cell completely safe to enter?

Trap: Confusing stopping with isolation. Stored energy, pneumatic pressure, gravity-loaded axes and other machines may remain; maintenance requires lockout/tagout.

RAA-498 · Is a safety function implemented in a standard PLC with redundant code safe?

Trap: Thinking duplicate software equals redundancy. Common-cause failures in one CPU or I/O card defeat duplicated code; certified safety PLCs or relays with diverse channels and diagnostics are required.

RAA-499 · Is a cobot safe if it stops on contact?

Trap: Ignoring clamping and tool hazards. A stop doesn't release a person pinned against a fixture, and sharp or hot tools injure even at low force. Contact must be within allowable limits and clamping situations designed out.

RAA-500 · If a robot cell passed a risk assessment, does it remain compliant forever?

Trap: Treating safety as a one-time certificate. Changes in tools, speeds, layout, programs or tasks can introduce new hazards, so the risk assessment and validation must be repeated after modifications.

Other levels: Basic · Intermediate · Advanced · All Robotics & Automation questions