Full-Year Summary

Year at a glance

The team began with a working Year 4-style MVP centered on the Intel server, an incomplete safety implementation, an early Qt HMI concept, and a three-LiDAR sensor arrangement with reliability problems. During fall semester, the team rebuilt the safety code as a finite-state machine, demonstrated stop-sign and path-following behavior, connected live data to the HMI, benchmarked perception, and evaluated FAST-LIO2 using indoor LiDAR data.

During winter, the architecture changed substantially. The team replaced LiDAR 3, introduced two Jetson AGX Orin computers for the new perception pipeline, developed Docker-based sensor interfaces, built a new HMI, added controlled GNSS attenuation, and rewrote much of perception for a distributed system. The first replacement LiDAR later failed, forcing another hardware decision and a second replacement.

Spring focused on integration and evidence. GPS reached the HMI through ROS 2, safety behavior ran in the vehicle, the new LiDAR produced data, perception connected to controls, and the team performed longer vehicle tests. Late testing exposed perception, mapping, thermal, and integration weaknesses. The vehicle passed technical and safety inspection on the first day of competition, and the team earned four static-event placements.

September 2025: scope and architecture

Week 01: September 8

Program setup

• The first Sunday meeting took place successfully.

• Leadership completed team-placement interviews and began onboarding members through NDAs and access procedures.

• A central roster consolidated contact information, membership, and attendance.

• The team reviewed the complete Year 5 event structure rather than treating each deliverable separately.

Competition requirements identified

• DYOC Run 1 required autonomy to resume after a manual takeover.

• DYOC Run 2 introduced degraded GNSS during the route.

• The Buy-Off Ride required an autonomous passenger mission with a visible route and rider interaction.

• The team also reviewed localization, Mobility Innovation, 0-0-0, the Concept Design Report, Concept Design Event, Simulation Challenge, SRS, and final safety deliverables.

• The final-year documentation needed to explain the vehicle's evolution across the full five-year program.

Week 02: September 22

Compute and electrical diagnosis

• The team identified a likely cause of the Intel machine shutdown: a poor 12 V battery-to-inverter connection caused by an incorrectly installed screw.

• The team scheduled another physical verification before deciding whether to replace the 12 V battery.

Safety-code redesign

• The Year 4 implementation had clear CAN receive/transmit code but lacked a proper finite-state machine.

• It used unnecessary processes and did not provide a clear controller-integration path.

• The Year 5 design adopted PyTransitions for a recurring state machine.

• The design used Python multiprocessing events and data classes.

• Steering and propulsion activation were targeted for October 3, followed by manual takeover and fault checking on October 10.

HMI concept

• The proposed interface used a rotated touchscreen mounted ahead of the factory GM infotainment screen.

• The team considered a custom printed enclosure, vent mounting, and physical input devices.

• Qt was selected for the GUI.

Week 03: September 29

Compute-platform commitment

• The team submitted an ETS request for two Jetson AGX Orin 64 GB computers for Perception.

• The request also included one Jetson Nano Super for Sensors, if available.

• This is the earliest explicit commitment in the supplied decks to the new distributed perception-compute platform.

Perception roadmap

• The team documented summer progress.

• Perception separated the existing MVP pipeline from future Jetson modules.

• The plan placed MVP demonstration testing ahead of the full Year 5 pipeline and DYOC integration.

• On-site data collection and end-to-end testing required a safety driver.

• The team proposed a notification workflow to recruit available safety drivers.

October 2025: first executable demonstrations

Week 04: October 6

Safety implementation

• CAN receive was reported at 95 percent.

• Safety logic was reported complete and implemented as a recurring finite-state machine.

• CAN transmit remained at 50 percent.

• The refined code was reported at 85 percent.

• Initial controller integration was underway. HMI and full-pipeline integration had not started.

• The recorded safety demonstration showed engagement checks executing and reporting completion.

Test discipline

• Leadership introduced progress tracking for SRS, safety evidence, and other competition deliverables.

• Future autonomy testing was expected to use the safety FSM, takeover handling, communications-fault handling, and blue-light behavior.

Week 05: October 13

Safety-requirement process

• Leadership established one requirement-lifecycle document per department.

• Each requirement needed deliverables and review comments.

• Weekly review times were assigned for Sensors, Controls, Safety Code, and other teams.

• Controls had 17 requirements remaining. Perception had 11, Sensors eight, and Hardware six.

HMI Mk1

• The team selected a 12.3-inch 1920×720 external display.

• An NVIDIA Jetson Nano drove the display because the interface required 3D rendering.

• Frontend and backend development proceeded in parallel using Qt.

• The team targeted a mid-November MVP.

• The recorded Mk1 demo showed the first working visual interface.

Pedestrian robot

• The hardware capstone team finalized an improved mobile pedestrian-robot design.

• The team expected the platform to support BYO/DYOC scenarios.

• Purchasing was underway, with completion projected for the end of November.

Week 06: October 20

Controls and vehicle behavior

• Two people began working on MPC.

• Controls continued validating lane-change behavior and defining completion and infeasibility conditions.

• The team discussed three-lane selection, maneuver cost, and oversteer limits.

• A camera failure delayed lane-change and perception testing by approximately one week.

• The recorded stop-sign test showed the behavior running on the vehicle in a closed lot.

Sensors and hardware

• The team prioritized camera recovery because both Controls and Perception depended on it.

• Hardware continued ingress-protection work for sensor housings and exposed roof connections.

• The Safety team planned to repair and test blue-light behavior by October 26.

Week 07: October 27

Perception campaign

• Perception planned ground-truth collection for barrels, pedestrians, signs, lights, and other competition objects at distances of at least 35 m.

• The next step was numerical evaluation against perception accuracy requirements.

• Controls requested better object-localization accuracy and work on three-lane scenarios.

Hardware and sensor reliability

• Hardware planned an Intel window and protection for exposed sensor connectors.

• GPS failures shifted sensor deadlines.

• Sensors continued timestamp serialization and latency tests.

Vehicle integration

• The team prioritized early HMI integration.

• Safety requirements were scheduled for HMI integration through November.

• The longer plan placed electrical-safety integration in November through February and final safety testing in February and March.

• Perception had to decide which 2D detector, 3D detector, pipeline, and SLAM implementation would run on the Jetsons.

November 2025: integration, sensor failure, and measured perception

Week 08: November 3

Requirements evidence

• Each technical group moved from requirement drafting toward evidence collection.

• Perception continued ground-truth and detection-accuracy work.

• Controls continued trajectory and vehicle-response tests.

• Sensors investigated timing and message-order problems.

• Hardware continued weather protection and integration work.

Week 09: November 10

HMI integration

• The touch display integrated successfully with its Jetson.

• On boot, the HMI Jetson pulled the latest scripts from GitHub, built the application, and launched it automatically.

• The application ran in Docker.

• Map and waypoint display worked.

• Backend integration with Controls, Perception, and Safety remained in progress.

• The recorded HMI demo showed map and live vehicle data.

Vehicle test

• The team recorded a path-following/yield-sign vehicle test.

LiDAR replacement trigger

• LiDAR 3 showed a measured delay greater than 17 seconds in one test.

• Perception observed approximately 500 lost LiDAR 3 messages out of 750.

• Camera and LiDAR timestamps also arrived out of order.

• Sensors explicitly requested purchasing new LiDAR hardware.

• Sensors and Perception were assigned to select a replacement by November 17.

Week 10: November 17

LiDAR alternatives

• VLP-16 and VLP-32 sensors were no longer available.

• The team reviewed Ouster OS1-32, OS2-32, and OS2-64 pricing.

• Perception and Controls discussed whether the vehicle needed one or two new LiDARs and whether rear coverage justified a second unit.

• Hardware began planning an elevated mount that remained below the competition height limit.

Jetson status

• The perception Jetsons were reported configured and reachable over SSH.

• Remote access to the Dell server was also repaired through the team VPN.

Controls and hardware

• Controls documented PID tuning and smoother velocity behavior.

• Hardware continued cable retention, sensor protection, and trunk-layout work.

Week 11: November 24

Perception benchmark

• The perception benchmark was approximately 90 percent complete, with missing frames still being repaired.

• The team labeled a two-barrel scenario as ground truth.

• The pipeline showed 2D detection, 3D fusion, and tracking in synchronized views.

• The team planned numerical evaluation across barrels, pedestrians, traffic signs, and vehicles at 10, 20, 30, and 40 m.

Localization model selection

• An image-based localization model had been tested and performed poorly.

• The team selected FAST-LIO2 as the LiDAR-based candidate.

• The recorded FAST-LIO2 runs used indoor LiDAR data.

• Testing with Buckeye vehicle/LiDAR data remained in progress.

• This checkpoint demonstrated algorithm feasibility, not final vehicle localization with the replacement LiDAR.

Sensors

• Sensors reported message serialization complete.

• Published average and maximum latency measurements covered three cameras, three LiDARs, and GPS.

• LiDAR 3 continued dropping messages despite improved latency results.

• The team planned a dedicated ROS bag and cable/switch-port swaps to separate sensor, cable, server, and bandwidth causes.

Hardware

• The replacement-LiDAR mount needed a defined position relative to GPS, a level orientation, acceptable vibration, adequate FOV, and compliance with the 88-inch height limit.

December 2025: validation freeze and new architecture plan

Week 12: December 1

Controls status

• Stop/yield logic, double lane change, a smoother velocity profile, PID improvements, and map-generation tooling were listed as implemented or tested.

• Varied lane-change conditions, emergency stopping, railroad crossing, rerouting, and some integration work remained open.

Perception status

• The deck repeated the November 24 perception recordings.

• These four embeds are exact duplicates and do not represent four new December tests.

• The old CPU/Dell MVP remained the near-term path for demos with Controls.

• The new Jetson pipeline remained under development.

New Jetson architecture

• The team selected a SLAM model and candidate 2D models.

• Work continued on 3D tracking and 3D detection.

• Perception and Sensors planned Docker-based sensor drivers.

• The February target included all models, communication with new sensors, and HMI start/stop integration.

Quantitative requirements

• Controls prepared lateral-error, velocity-error, reaction-time, and progressive-braking tests.

• Perception continued IoU and localization-error evaluation.

• Sensors investigated dropped LiDAR 3 messages.

Week 13: December 8

LiDAR replacement formally scheduled

• The deck explicitly listed “Replacing LiDAR 3.”

• The team planned to determine the arrangement of P90 and X90 units.

• Hardware would mount the replacement.

• Sensors would configure transforms, PTP time synchronization, and field of view.

Safety and diagnostics

• Sensors implemented frozen-sensor detection based on timestamps.

• Fault states needed publication to an HMI-readable ROS topic.

• Controls repeated reaction-time testing because recording did not start at the correct time in the previous test.

• Perception planned broader 2D/3D testing across object types and distances.

Testing schedule

• TRC visits were scheduled for January 25, February 8, and March 8.

Jetson integration risks

• Hardware raised concerns about Jetson heat inside the trunk.

• Jetson mounts and revised PCB-box placement remained prerequisites for wiring inspection.

January 2026: replacement hardware and safety re-baseline

Week 14: January 19

Replacement LiDAR identified

• The exact change was documented as “Replacing LiDAR 3, P90 to X90.”

• The January 26 target included time synchronization, FOV and blind-spot evaluation, transforms, and PTP configuration.

• Hardware needed to reroute the replacement cable through the roof opening.

TRC priorities

• Tier 0 required GPS route points and a minimal camera/GPS sensor stack, with LiDAR included if available.

• Tier 1 covered signage, pedestrian, lane-line, and object-placement recordings.

• Tier 2 covered new-LiDAR data and joint Perception/Controls checks.

Perception

• The January-February plan called for a Year 5 MVP pipeline and a new benchmark built around the replacement LiDAR.

Week 15: January 26

Requirements status

• Sensors reported 75 percent of validation complete and continued camera/LiDAR recalibration and watchdog development.

• Perception reported 20 percent complete. It needed new data from recalibrated sensors before validating detection and localization.

• Vehicle Integration reported zero percent of its small requirement set because the HMI remained unfinished.

• Hardware reported 90 percent complete, with the LiDAR mount, Jetson mounts, and 12 V monitoring still relevant.

• Controls reported 70 percent complete and expected two weeks of MVP work before resuming validation.

Safety analyses

• The team updated PHA, FIA, SEFA, and SOTIF material for the new perception pipeline, P90-to-X90 replacement, 360-degree LiDAR, Jetsons, HMI, watchdog, cooling, trunk layout, ROS network, and safety code.

February 2026: final hardware and distributed pipeline

Week 16: February 2

Assigned critical work

• Sensors owned GNSS attenuation, calibration, and new-LiDAR installation.

• Hardware owned the new mount.

• Perception owned the new pipeline, 2D/3D detection, SLAM, and integration of two Jetsons.

• Vehicle Integration owned HMI hardware and interface work.

• Safety owned watchdogs, architecture maps, and safety-code/VI requirement testing.

HMI and physical integration

• MVP dashboard hardware integration was complete.

• The team continued integrating Safety.

• The HMI mount went through one or two prototypes per week.

GNSS attenuation design

• The PwrPak7 used two GNSS antennas, so the vehicle required two attenuators.

• The existing attenuator blocked antenna bias power and lacked guaranteed performance across the receiver bands.

• The proposed replacement covered 1.1–1.7 GHz, passed DC power, and provided adjustable attenuation.

LiDAR mount risk

• Parts had been ordered.

• The proposed height created an approximately 2.5 m front blind spot and potential interference with the front rack and V2X/GPS equipment.

Week 17: February 9

HMI expansion

• The HMI included a data logger and uploader, embedded terminal, persistent settings, light mode, status indicators, FSM state, and ROS camera streams.

• Milestones included dashboard mounting, destination selection, autonomy engagement, route annotation, network configuration, saved backup credentials, and customization.

Two-Jetson MVP

• The Year 5 perception MVP was explicitly defined around two Jetsons.

• The pipeline needed training on DYOC objects, end-to-end testing at TRC, and integration with Controls.

• Perception proposed Zenoh and ROS 2 DDS Discovery Server for reliable Jetson-to-Intel communication.

Safety

• Safety-code integration with HMI and Controls continued.

• ROS 2 compilation problems remained an immediate blocker.

Week 18: February 16

New LiDAR installed and exercised

• Hardware mounted the replacement LiDAR with side supports.

• The cameras moved to the side and required FOV checks for calibration.

• The team planned to remove the Cepton and move a camera into its prior position.

• Perception collected new data with the replacement LiDAR for SLAM.

• A long recording showed preliminary 3D detection inference using pretrained models.

GNSS attenuation

• Sensors installed the attenuator with new SMA cabling.

• The team verified that it reduced the GPS signal and planned to plot position error at different attenuation settings.

Power and thermal planning

• Hardware began detachable Jetson-mount design.

• The vehicle still required a combined two-Jetson plus Intel power and heat stress test before DYOC.

Week 19: February 23

Safety behavior

• All safety checks were reported complete.

• The end-to-end GM state machine was developed.

• Manual takeovers had been tested.

• Blue-light behavior had been tested and worked.

• HMI integration was reported at 85 percent.

• Controller work was reported at 90 percent but still required testing.

Perception

• The team collected DYOC data.

• Remaining work included auto-labeling, model training, Jetson deployment, and end-to-end testing with the vehicle.

Program results

• The 0-0-0 Challenge was complete.

• Judges gave positive feedback on the Innovation Challenge and advanced the team toward the final competition.

March 2026: system interfaces and LiDAR failure

Week 20: March 2

HMI and ROS 2 integration

• The HMI consumed the Sensors team's live GPS ROS 2 topic.

• The recorded interface showed vehicle location updating on the cockpit map.

• The team continued the physical dashboard installation.

Perception and planning visualization

• A recorded right-turn scenario displayed vehicle pose and candidate path information.

• The recording made localization/planning behavior inspectable but did not by itself prove end-to-end vehicle control.

Week 21: March 9

Replacement-LiDAR failure

• A damaged connector interrupted calibration and new-data collection.

• Leadership considered sending the LiDAR to the manufacturer, buying another identical unit, purchasing through a third party, borrowing from another team, or completing DYOC without LiDAR.

• A replacement purchased through the official channel would consume almost the entire remaining major-parts budget and likely miss the DYOC deadline.

• Perception documented fallback options using HD-map geometry and image-based 2D detection.

• The team preferred retaining an Ouster sensor because most models and collected data targeted that LiDAR family.

Jetson status

• Sensors installed all drivers and dependencies on the 2D Jetson.

• The 2D Jetson could start the complete sensor suite.

• The team planned to repeat the setup on the 3D Jetson.

• Camera calibration improvements paused because of the damaged LiDAR connector.

Week 22: March 23

Safety case

• The final safety case was due in eight days.

• The team rebuilt the PHA and Safety Concept documents over the break.

• Safety began compiling validation evidence and planned final reviews with every technical team.

LiDAR contingency

• The deck stated that DYOC might proceed without LiDAR.

• Hardware questioned whether a replacement needed installation and cable routing during the same weekend.

• All Jetson mounts and resized PCB boxes were complete.

Perception testing

• The team prepared traffic signs and portable stands for validation and course setup.

Week 23: March 30

Transition to April execution

• The safety submission moved into final review and delivery.

• Remaining technical effort concentrated on replacement hardware, perception-to-controls integration, and vehicle testing.

• The team entered April with little schedule margin for another major architecture change.

April 2026: integrated vehicle stack

Week 24: April 6

Integrated HMI behavior

• The recorded cockpit test showed HMI, Safety, and Controls working together.

• Vehicle state and AV controls appeared in the interface.

• The test included brake activation.

Second replacement LiDAR

• A LiDAR sourced through eBay passed an initial bench/network test.

• The team still needed waterproof cabling and final installation.

• PTP clock synchronization was repaired.

• The team had not received a response from the original replacement-LiDAR supplier.

Integration schedule

• Hardware planned Ethernet and power routing through the roof.

• The team planned a permanent HMI mount, BlueLight-controller placement, and in-trim wiring.

• Perception planned 3D detector integration followed by SLAM integration and vehicle testing with Controls.

Week 25: April 13

Schedule pressure

• The 3D detector integration deadline moved from April 12 to April 19.

• LiDAR-based SLAM remained the highest-priority uncertain item.

• Perception expected 2D/3D data for Controls after April 19 and SLAM after April 26.

• The team scheduled HMI engagement-condition display for April 17 and hazard-light control for April 24.

Physical installation

• Roof routing, LiDAR mounting, trunk retention, cable organization, and summer-temperature stress testing remained active.

Week 26: April 20

Perception MVP

• Perception reported an executable MVP pipeline.

• It provided the objects Controls required for vehicle behaviors.

• Crash avoidance and 2D detection were marked complete.

• HD-map lane information remained incomplete.

• The team's main priority shifted to vehicle testing.

End-to-end test plan

• The team planned full tests from Sensors through Perception, Controls, Safety, and HMI.

• Finals delayed active testing until the end of the first week of May.

• The plan allowed roughly three and a half weeks of final testing.

Hardware completion

• HMI wiring was routed under the interior trim.

• A new outlet provided USB-C and USB-A power for the BlueLight controller and cellular hotspot.

• Hardware hid the BlueLight controller under the carpet.

• A printed roof cover and sealant addressed the new cable opening.

May 2026: final testing and competition preparation

Week 27: May 4

Vehicle testing

• The team recorded a low-light vehicle run with visible in-cabin controls.

• This is the final unique Buckeye demonstration embedded in the supplied decks.

• Vehicle Integration and Controls continued full-system operation and longer test sessions.

• Thermal endurance remained important because the trunk contained two Jetsons plus the Intel computer.

Media distinction

• A second video on the same slide was an external comparison/reveal clip.

• It should remain labeled as reference media, not Buckeye test evidence.

Week 28: May 13

Perception telemetry

• Perception produced usable telemetry from vehicle operation.

• Plots and debug views exposed missed detections, false detections, localization error, and unstable cases.

• The team could now prioritize failures using recorded evidence instead of driver impressions alone.

• The “Perception works… almost” slides documented that successful frames did not yet guarantee robust continuous behavior.

Week 29: May 20

Competition scenario rehearsal

• The team documented physical setups for Demo 1 and Demo 2.

• Course assets included barrels, signs, and Type III barricades.

• The rehearsal aligned object placement, route flow, expected behavior, and team roles.

• These images document preparation and test configuration. The deck does not contain a new embedded vehicle video for this checkpoint.

Week 30: May 27

Competition freeze

• The final checkpoint concentrated on configuration stability rather than new features.

• The team needed to preserve known-good software, logs, recovery procedures, and hardware configuration.

• Operators and inspection responsibilities needed clear ownership before travel and scoring.

• The supplied section contains no new unique embedded demonstration.

June 2026: competition outcome and lessons

Week 31: June 9

Competition result

• The vehicle passed technical and safety inspection on the first day.

• The team placed second in the Simulation Challenge.

• The team placed third in Mobility Innovation.

• The team placed third in Safety Reports.

• The team placed third overall in static events.

Controls reflection

• Motion planning connected perception and the HD map to the controller.

• MPC generally worked.

• Every major code path received at least some exercise during the scoring run.

• The HD map contained errors and did not suit the controller well.

• Behavior planning did not generalize reliably at Mcity.

• MPC performance remained limited by upstream information and integration quality.

Perception reflection

• The new distributed architecture used Jetson Orin computers effectively.

• The architecture change required a near-complete module rewrite.

• Latency forced additional C++ rewrites of heavy modules.

• Late-stage, hand-tuned fusion remained brittle and produced missed detections, misclassifications, and false detections.

• Map localization worked with FAST-LIO2 as a fallback.

• The team lacked a sufficiently complete test pipeline to guarantee overall quality before competition.

Sensors reflection

• Sensors rebuilt calibration processes for the changed camera and LiDAR geometry.

• The first new LiDAR failed.

• The team helped integrate a second new LiDAR.

• Camera position and heading changed multiple times, requiring new intrinsic and extrinsic calibration.

• The GNSS attenuator supported degraded-localization testing.

• The V2X challenge was cancelled, so the vehicle did not use V2X in competition.

Hardware and integration reflection

• The final system included major trunk, mounting, cooling, roof-routing, and cable-management changes.

• Repeated physical changes increased calibration and integration work for other subteams.

• The final retrospective identified insufficient end-to-end test time as a central program weakness.