Vandalism Resistant Automated Speed Enforcement System
A compact, elevated speed-enforcement prototype combining Doppler radar, deterministic control, adaptive illumination, and stereo computer vision.
FINAL PROTOTYPE · FRONT ASSEMBLY
20 mvalidated radar range
50.3 msradar sampling interval
<31.5 µsambient-light response
±2 m/smeasured speed accuracy
01 / CHALLENGE
Move enforcement hardware out of reach.
Conventional roadside speed cameras are vulnerable to vandalism. The team’s design goal was to reduce system size and weight enough for installation on existing elevated infrastructure—without sacrificing capture timing, night operation, or multi-target identification.
02 / ARCHITECTURE
Two sensing paths. One verified event.
An HB100 Doppler sensor measures approach speed while an Arduino Mega 2560 executes the real-time control schedule. A ZED 2 stereo camera and computer-vision pipeline track multiple objects, estimate depth-based velocity, and match the correct target to the radar measurement.
FUNCTIONAL SYSTEM FLOW
HB100 SIGNAL CONDITIONING
03 / ELECTRONICS
From a weak radar return to a clean digital signal.
The raw HB100 output was too small for direct microcontroller input. A custom two-stage active band-pass filter amplified frequencies from 152 Hz to 4.34 kHz, rejected noise, and used an analog comparator to produce a 0–5 V square wave.
After Multisim validation, the team assembled the custom PCB by hand and confirmed reliable frequency counting and speed measurement.
04 / REAL-TIME CONTROL
Deterministic timing under a moving target.
The report distinguishes the 50.3 ms radar sampling interval from shorter calculation and output operations. A 250 ms frame-based schedule kept sensing and control predictable while image processing ran on a separate subsystem.
T1Radar sampling50.3 ms
T2Speed calculation183 µs measured
T3Ambient-light reaction<31.5 µs
T4Flash trigger delay0.42 µs
05 / VALIDATION
Measured, not assumed.
REAL-WORLD SPEED TEST
Logic-analyser and real-world tests verified the timing model. The prototype reacted to ambient-light changes in under 31.5 μs, sampled radar speed in 50.3 ms, and consistently measured moving-object speed within 2 m/s.
The project began with a client living with cerebral palsy and limited upper-limb mobility. Interviews, a modified Barthel Index survey, and occupational-therapy feedback identified the tasks with the greatest impact: leaving home independently, operating lights and elevator buttons, and using access fobs—all through a control method requiring minimal physical and cognitive effort.
02 / SYSTEM CONCEPT
One device. Three coordinated systems.
The solution combines a removable wheelchair mount, a multi-joint manipulator with modular end effector, and a low-effort control and feedback interface. Safety, comfort, appearance, vibration, reach, and caregiver access were treated as engineering requirements rather than afterthoughts.
WHEELCHAIR INTEGRATION CONCEPT
FINAL ARM · CAD VIEWS
03 / PROTOTYPE VS. FINAL DESIGN
Proof of motion, then a stronger architecture.
The early functional prototype used four degrees of freedom and a stylus to validate autonomous button actuation. The final system moved to six serially controlled servos, 6061 aluminum links, a removable armrest mount, and a distributed compute architecture.
Low-cost prototype servos jittered and lacked torque at full extension. Those findings drove the stronger mechanical design and the move to UART servo control in the final arm.
04 / HUMAN–MACHINE INTERFACE
Clear feedback with minimal effort.
The interface gives the user a direct view of the robot’s available actions and current state. Its task-oriented interaction model reduces the number of physical inputs and avoids forcing the user to manually coordinate individual joints.
The video demonstrates the working HMI and its role in making autonomous functions understandable and predictable.
HMI DEMONSTRATION · SELECT PLAY TO VIEW
05 / PERCEPTION & CONTROL
High-level perception. Deterministic motion.
The final architecture used an Intel RealSense depth camera with a Raspberry Pi 5 and AI HAT for perception and high-level decision-making. An Arduino Sense controller handled the real-time control layer, while the six servos received position commands over a serial UART bus. This separation kept compute-intensive vision away from timing-sensitive motion control.
The earlier demonstration shown here used a ZED 2, AprilTags, a Jetson Nano, and an Arduino-based servo controller to validate autonomous elevator-button targeting. A hard-wired emergency stop provided an immediate physical shutdown during testing.
EARLY PROTOTYPE · INTEGRATED VISION DEMONSTRATIONEARLY PROTOTYPE · APRILTAG POSE TEST
06 / PROTOTYPE & LEARNING
A working concept—with useful failures.
EXPLODED ASSEMBLY & BILL OF MATERIALS
Power and vision modules passed their initial tests, including accurate AprilTag localization. Low-cost prototype servos introduced torque limitations and jitter at full extension, but calibration enabled the integrated arm to autonomously detect and actuate the target buttons—validating the overall system logic.
User needs translated into weighted engineering requirements
Emergency stop integrated into the physical prototype
Torque and jitter findings carried into the final design
TOOLS & TECHNOLOGIES
Raspberry Pi 5 · AI HAT · Intel RealSense · Arduino Sense · UART servo control · Python · OpenCV · inverse kinematics · SolidWorks · FEA · DFM · FMEA
CASE STUDY 03MECHATRONICS DESIGN · THREE-PHASE TEAM PROJECT
Autonomous Package Delivery Robot
One mobile robot iterated across three escalating challenges—from rapid terrain traversal to multi-package navigation inside a structured maze.
PHASE C CONFIGURATION · FINAL INTEGRATED CAD
3design phases
0.57 m/svalidated flat speed
1.5°average turn error
150 msvision response
01 / DESIGN BRIEF
Keep the platform. Raise the challenge.
The APDR was intentionally developed as one continuous system. Each phase retained the four-wheel platform and closed-loop drive while adding new terrain, manipulation, sensing, and navigation demands. Parametric SolidWorks models, FDM manufacturing, modular connectors, and shared mechanical interfaces made rapid redesign practical between demonstrations.
PHASE AMobility
Traverse and deliver
Cross grass, pebble, and mulch; drive onto the target platform; and release a package using encoder-based PID control.
ResultFastest terrain run at 6 seconds; package landed 2.5 cm outside the target.
PHASE BNavigation
Descend, search, retrieve
Add ramp travel, wall-following, maze entry, ToF sensing, and a pulley-driven lifting gripper to retrieve a block.
ResultReached the maze entrance, but hardware failures and false wall detection prevented entry.
PHASE CIntegration
Plan, align, and recover
Integrate IMU and ToF fusion, ESP32 vision, A* path planning, PD correction, revised electronics, and two-package handling.
ResultDelivered the first package and reached the second; accumulated localization error prevented final recovery.
02 / CONTROL EVOLUTION
From fixed motion to sensor-informed autonomy.
Phase A used encoder feedback and per-wheel PID loops for straight driving and in-place turns. Phase B introduced event-driven wall following with distance sensors. Phase C combined IMU heading, filtered ToF measurements, PD corrections, vision-assisted package alignment, and an A*-generated route through the known maze.
UPPER TERRAIN & LOWER MAZE ENVIRONMENT
REVISED POWER & SIGNAL DISTRIBUTION
03 / HARDWARE ITERATION
Failures became design inputs.
Wiring noise and last-minute hardware faults in Phase B directly shaped the Phase C rebuild. The team revised the distribution boards, separated power and signal paths, standardized JST connections, improved wire management, and added serviceable sensor mounts.
The mechanical platform evolved in parallel with a rack-and-pinion gripper, pulley lift, ESP32 camera mount, and lightweight printed geometry designed around ±0.2 mm manufacturing tolerances.
04 / VALIDATION
Test at the component, subsystem, and mission levels.
PHASE C MISSION FLOW
Three levels of testing separated sensor and actuator accuracy from navigation performance and full-mission behavior. This made failures traceable and quantified the gains from the final sensor-fusion architecture.
ToF sensing measured within ±1 cm
Encoder error remained below 1.42%
IMU-assisted 90° turns averaged 1.5° error
Vision centered the gripper within ±1 cm in 8 of 10 trials
Grasp and release succeeded in five consecutive tests
05 / ENGINEERING TAKEAWAY
Phase B was not a dead end—it revealed exactly what the final system lacked.
The three-phase structure made reliability visible. The final robot integrated more capable mechanics, electronics, and software than the initial platform, but the official run also exposed how small heading and distance errors accumulate across long autonomous sequences. The project closed with a clear next step: stronger localization and longer-duration sensor-fusion testing.
A 2024 academic design study integrating power electronics, two-axis actuation, and closed-loop control for a simulated 100 kg tracking platform.
FINAL TUNED SIMULATION · TARGET AND TRACKING RESPONSES
100 kgmodeled platform load
15 kWcombined motor rating
600 kHzboost switching frequency
±2°tracking result within 4 s
01 / SCOPE
Integrated simulation of a high-load tracking system.
The course project modeled the interaction between electrical power delivery, azimuth and elevation actuators, and a feedback controller. Requirements included supporting a 100 kg load, acquiring feasible target trajectories, and keeping the target near the camera centre. The submitted work was a design and simulation study rather than a fabricated turret.
02 / POWER ELECTRONICS
Battery supply and high-voltage conversion.
Six 25.4 V, 166 Ah battery models were arranged as three parallel strings of two series-connected batteries, producing a modeled 50.8 V, 498 Ah source. A boost-converter stage was sized around an 83.7% duty cycle and 600 kHz switching frequency, with calculated inductance, capacitance, ripple, and semiconductor requirements.
MODELED BATTERY CONFIGURATION · 50.8 V, 498 AH
DC-DC BOOST-CONVERTER MODEL
03 / COMPONENT SIZING
Conversion performance derived from operating requirements.
The design calculated a 1.67 µs switching period, 1.39 µs pulse width, 9.6 A allowable current ripple, and 3.11 V allowable voltage ripple. The resulting model used a 1,920 µH inductor and 13 µF output capacitor, followed by component selection against the simulated electrical stresses.
Early full-detail Simscape models produced algebraic-loop issues and simulation times ranging from more than 30 minutes to over an hour. The final architecture represented the BLDC motor and power stage with reduced-order transfer functions, cutting execution to milliseconds and allowing MATLAB PID Tuner to be used iteratively.
FINAL REDUCED-ORDER TWO-AXIS CONTROL MODEL
05 / RESULT
Final response evaluated against a moving target.
PROJECTILE-MOTION INPUT · TUNED RESPONSE
The final 20-second simulation included motor parameters, load, Gaussian measurement noise, a 5.6 kHz low-pass filter, and tuned PID control. The report records the simulated response as reaching within ±2° of the moving target within four seconds.
Two-axis position feedback modeled
Noise added before feedback-error calculation
Power-stage response included as a first-order model
Result reported as simulation, not physical validation
The case studies document requirements, development decisions, testing, and observed results.
01 / REQUIREMENTS
Define the problem
Identify user needs, operating constraints, performance targets, and technical risks.
02 / DEVELOPMENT
Design and prototype
Compare concepts, model key behavior, build subsystems, and integrate the selected design.
03 / VALIDATION
Test and document
Measure performance, record failures, implement revisions, and report the final results.
03
Engineering experience
Manufacturing, qualification testing, laboratory operations, and controlled technical documentation.
2024–2025Peterborough, Ontario
Manufacturing Engineering Intern
BWXT Nuclear Energy Canada
Coordinated qualification testing of nuclear motor assemblies for Ontario Power Generation under controlled contaminated-material conditions.
Authored routing packages, process documentation, test procedures, and production instructions within a Canadian nuclear quality-management environment.
Set up, maintained, and troubleshot undergraduate engineering laboratory equipment.
Designed, assembled, and tested an interactive embedded/mechanical model analogous to CANDU reactor-control fundamentals.
Prepared operating procedures and equipment instructions and supported expansion of the student design studio.
04
Publication
Conference work related to the educational reactor simulator developed at Ontario Tech University.
CONFERENCE PAPERJUNE 2025
Development of a CANDU Reactor Simulator for Educational Applications
A. Machrafi, A. Bendali-Braham, E. Guenette, E. Kengonzi, I. Voloshin, A. Banerjee, J. Morrison, T. Tan, H. Bachu, R. Yau, P. Marchut, A. Quevedo, and S. Perera
44th Annual CNS Conference and 49th CNS/CNA Student Conference · Toronto, Ontario · June 8–11, 2025
The paper documents a physical CANDU educational simulator that combines adjustable control rods, precomputed OpenMC neutron-flux distributions, an Arduino interface, and an LED lightboard to make reactor behavior visible during laboratory instruction.
Peter Marchut is a Mechatronics Engineering graduate from Ontario Tech University with project experience across mechanical design, embedded control, instrumentation, automation, and system testing.
His engineering experience includes nuclear manufacturing, undergraduate laboratory operations, multidisciplinary robotics, qualification testing, technical documentation, and hands-on troubleshooting.
Qualification testing, troubleshooting, test procedures, technical reports, routing packages
Supporting software
Python, Git, Linux
SELECTED COURSEWORK
Real-time embedded systems · automatic control · industrial automation · sensors and instrumentation · robotics and automation · actuators and power electronics · CAD · microprocessors and digital systems
EMBEDDED SYSTEMS · TEAM PROJECT
CAPSTONE · THIRD PLACE
MECHATRONICS · THREE PHASES
SYSTEM MODELING · TEAM PROJECT
