Back to Selected Work

Physical AI · 2026

RubbleEcho

A rover and operator console for prioritizing inspection in a controlled rubble testbed.

Team Lead · McNeeseNexus

Nebius × NVIDIA Global AI Hackathon

I brought the hackathon opportunity to McNeese, pitched recruitment through ACM, led the team, and worked on rover assembly and system integration.

Watch the demoView source
McNeeseNexus · Assembled rover

Demo

The submission video shows the rover and its sensor hardware, the controlled testbed, and the dashboard used to review and export a trial. The system notes below follow the final build documentation for the acoustic workflow.

Watch on YouTube2 min 41 sec · Hackathon submission

What we built

We wanted to give an operator a way to compare signals from different spots before choosing where to inspect. We built a controlled sand-table testbed, a rover that collects evidence by square, and a dashboard that turns those readings into a search-priority map.

Fixed sweep across the test grid
9 cells
Outer footprint of the Rev C sand table
105 × 80 cm
First-cell listening interval in the documented sweep
12 seconds

Tech stack

AI inference
  • Nebius Token Factory
  • NVIDIA Nemotron

Nemotron 3 Nano reviews the sensor evidence. Nemotron 3.5 Super VL describes camera frames when available.

Dashboard
  • React
  • TypeScript
  • Vite
  • Tailwind CSS
  • Framer Motion
API and signals
  • Python
  • FastAPI
  • Uvicorn
  • Pydantic
  • HTTPX
  • NumPy
  • SciPy
Rover and sensors
  • Raspberry Pi 5
  • Arduino Uno
  • Seeed Studio
  • GPIO Zero
  • pySerial
  • ALSA (arecord)
Testing and lint
  • pytest
  • Playwright
  • ESLint
System diagram

Motion and control

  1. Phone remote
  2. Laptop command queue
  3. Raspberry Pi
  4. Arduino Uno and motors

The operator can drive manually or arm a fixed nine-cell sweep in the firmware.

Sensing and review

  1. Ultrasonic range, 60 GHz radar, audio, and an available camera frame
  2. Raspberry Pi
  3. Laptop dashboard and evidence store
  4. NVIDIA Nemotron on Nebius Token Factory
  5. Heatmap and operator recommendation

The laptop classifies recorded audio windows. After the sweep, Nemotron reviews the stored evidence. The vision model describes a cell only when a camera frame is available.

The operator uses the heatmap to review the readings for a square and decide where to inspect next. The motor queue carries commands from the operator, and the Arduino runs the armed sweep. The model returns its recommendation to the dashboard for the operator to review.

Read the final technical documentation

Recruitment and build journal

From the ACM pitch to the build team

I brought the hackathon opportunity to McNeese ACM and pitched a recruitment plan at the first chapter meeting of the semester. We followed up with a Luma sign-up window for McNeese engineering students from September 23–28, 2026. That gave interested students a way to join the team before the build sprint.

  1. 01

    ACM pitch

    First chapter meeting of the semester

  2. 02

    Luma sign-ups

    September 23–28, 2026

  3. 03

    McNeeseNexus

    The build team comes together

  4. 04

    Build sprint

    From concept to the full rover

Opportunity

Lime green Nebius and NVIDIA Global AI Hackathon artwork with both company logos.

01 / Opportunity

Hackathon artwork

Official artwork for the Nebius × NVIDIA Global AI Hackathon.

Cropped Devpost email welcoming Prince to the Nebius × NVIDIA Global AI Hackathon.

02 / Opportunity

Acceptance confirmation

Devpost confirmed my acceptance into the hackathon.

Recruitment and concept

Kody Vo stands by a projected slide as chapter members sit around the lab tables.

03 / Recruitment

First ACM chapter meeting

I pitched the recruitment plan at the first ACM meeting of the semester. Former chapter president Kody Vo is speaking in this photo.

Ten storyboard panels show a concept rover, a rubble testbed, acoustic signals, and proposed inspection maps.

04 / Concept

Pre-sprint storyboard

Pre-sprint concept illustration showing how the rover might gather signals and guide the next inspection. The panels depict a possible knock-and-listen workflow and compare hypothetical search strategies. The illustrated results are design examples.

Lab work

Rajesh Karki holds a breadboard beside a monitor and wired electronics.

05 / Lab work

Sensor testing

Rajesh Karki tested the sensors and checked that the hardware responded.

Multimeter probes touch a Raspberry Pi on the workbench beside a breadboard and keyboard.

06 / Lab work

Raspberry Pi test bench

Cooper Tadlock worked on the Raspberry Pi at the test bench.

Dipendra Kandel solders parts on the wheeled chassis with a notebook and tools beside him.

07 / Lab work

Rover assembly

Dipendra Kandel assembled the rover and checked how it moved.

Prince is in the foreground as Dipendra works beside the rover parts and soldering stand.

08 / Lab work

Overnight assembly

Dipendra and I worked on the assembly from around midnight into the early morning.

Two people sit next to the partially assembled rover and wiring in the lab.

09 / Lab work

An ACM alumnus visits

A former McNeese ACM leader returned to the lab to help and check in with the team.

Dr. Vipin Menon stands by the lab table as teammates work at a computer and whiteboard.

10 / Lab work

Advisor lab visit

Dr. Vipin Menon, our ACM advisor, stopped by to see how the build was going.

Outside the lab

I invited teammates and ACM members to celebrate Dashain with the Nepali Student Association at McNeese.

A person in pink applies tika to Caden Fontenot beside flowers and a ceremonial table.

11 / Dashain

Tika for Caden

Caden Fontenot receives tika and a blessing at the Nepali Student Association’s Dashain celebration at McNeese.

Cooper Tadlock holds his hair back beside the Dashain ceremonial table.

12 / Dashain

Tika for Cooper

Cooper Tadlock receives tika and a blessing at the celebration.

Evan Weber leans in to receive tika beside a table with flowers and offerings.

13 / Dashain

Tika for Evan

Evan Weber, McNeese ACM’s treasurer, receives tika and a blessing.

Three people wearing tika sit together with plates of food.

14 / Dashain

A meal together

We shared Nepali food after the Dashain celebration.

Hardware

The Seeed Studio box identifies the ReSpeaker XVF3800 USB four-microphone array.

15 / Hardware

ReSpeaker microphone array

The Seeed Studio ReSpeaker XVF3800 four-microphone array used in acoustic testing.

Seeed Studio packaging shows the 60 GHz mmWave sensor enclosure and specifications.

16 / Hardware

60 GHz radar

The Seeed Studio 60 GHz mmWave radar module used in the project.

From first prototype to full rover

An overhead view shows the early wooden chassis with yellow wheels, batteries, a motor driver, and exposed wiring.

17 / Prototype progression

First working prototype

An overhead view of the early chassis, wiring, and control electronics.

The assembled four-wheel rover faces forward with its twin ultrasonic transducers, cabling, and onboard boards visible.

18 / Prototype progression

Full rover

The assembled RubbleEcho rover with its front sensors and onboard electronics.

McNeeseNexus

Team and contributions

Prince Pudasaini
Team lead · Recruitment, coordination, rover assembly, and system integration
Dipendra Kandel
Co-leader · Hardware selection, mechanical design, and rover assembly
Arjun Gautam
Digital design and testbed layout
Pratik Thapa
Testbed construction
Cooper Tadlock
Rover 3D printing, Raspberry Pi setup, and sensor integration
Caden Fontenot
GitHub repository setup and code integration
Pratikshya Parajuli
Dashboard development
Rajesh Karki
Sensor testing and demo video
Yuni Lin
Dashboard development

Advisors and community

Thanks to ACM advisor Dr. Vipin Menon, former chapter president Kody Vo, visiting ACM alumni, ACM treasurer Evan Weber, and the McNeese ACM and Nepali Student Association communities.

Submission and source