20 réponses à ce sujet

[vishal095015]

 d.jpeg   114,06 Ko   0 téléchargement(s)
 c.jpeg   167,9 Ko   0 téléchargement(s)
 a.jpeg   114,06 Ko   0 téléchargement(s)

[vishal095015]

After my first AI Vision project, I went deeper. This time I built a fully autonomous voice-controlled robot that thinks, speaks, sees, and moves - all by itself!
About Me:

Vishal Bharti, ECE student at Dr. Harisingh Gour University, Sagar MP. YouTube: Irisrobonium
The Big Idea:

A plain English file controls the entire robot. No hardcoded logic. The AI reads instructions, decides what to do, and acts physically. Change 3 lines of text - robot changes its entire personality.

[vishal095015]

Hardware:

RDK X5 (10 TOPS BPU) - AI brain
Raspberry Pi Pico 2W - motor controller
L298N - motor driver
HC-SR04 - obstacle detection
OPPO F27 - live camera via WiFi
3.5mm mic + speaker - voice I/O

Software:

Codex CLI (GPT-5.5) - agentic AI brain
AGENTS.md - plain English robot instructions
MicroPython - Pico 2W firmware
Whisper STT - voice input
Piper TTS - voice output
YOLOv8 on BPU - object detection

[vishal095015]

Logical Thinking Brain:

IrisBot doesn't follow hardcoded commands - it reasons through goals like a human. When I said "move forward for 1 second" here's what it actually did internally - completely on its own:
Goal: move forward 1 second
-> "Check distance first" (safety rule)
-> Distance = 13.7cm -> TOO CLOSE -> refuse to move
-> Recheck -> Distance = 252.5cm -> CLEAR
-> Move 0.5s -> recheck -> 211.3cm -> still safe
-> Move 0.5s more -> total 1 second -> stop -> report
Nobody programmed this sequence. No if/else. GPT-5.5 reasoned every step itself using 5 types of logic - Safety, Causal, Planning, Recovery and Memory logic.

[vishal095015]

What it can do:

Say "move forward" -> robot checks distance -> moves safely
Say "what do you see" -> camera -> GPT-5.5 describes the scene
Obstacle detected at 13.7cm -> robot stopped automatically - zero hardcoded logic!
Full voice conversation loop - speak, it listens, thinks, acts, replies

Coolest part?

The robot checked distance before moving - found obstacle at 13.7cm, refused to move. Cleared path at 252.5cm - moved autonomously. I never wrote a single if/else for that. The AI decided itself!
Next Level Plans:

Person following using BPU YOLOv8
SLAM room mapping
Offline AI fallback using Llama 3.2 1B
Multi-robot fleet communication

Built with on D-Robotics RDK X5 | Dream Keeper Challenge 2026

[drobo_community]

Hi @vishal great work, please update it in the main thread. @Shockley please leave your comments.

[shockley6668]

Hi @vishal , where is your repo?

[vishal095015]

Please give me some time I will upload everything

[vishal095015]

https://github.com/v...t-Agentic-Robot

[shockley6668]

Hi, @vishal , great concept, but Stage 2 needs a real proposal document: add Concept/Application Design, AI System Architecture with diagrams and compute/ROS details, plus Engineering Plan with BOM, roadmap, risks, and project structure.

[vishal095015]

Yes please help me for this all I need guidance, for what exactly I have to create

[vishal095015]

Still I am working on this project

[vishal095015]

And it's my humble request please help me

[shockley6668]

You can follow this guide:https://github.com/D-Robotics/Robotics-Dream-Keeper-Challenge/blob/develop/stages/stage2-build.md

[vishal095015]

Thank you very much and please give me some time I will definitely do exactly this all

[vishal095015]

Sahayak - IrisBot GuardianAI | Vishal Bharti (Stage 2 + Stage 3) completed

[vishal095015]

**Sahayak - IrisBot GuardianAI** - an offline-first, voice-controlled autonomous guardian robot on the RDK X5, built for elders and home care. It sees, listens, thinks, moves, and speaks - all core AI on the BPU, cloud optional.

**Deliverables**
Demo video:
Repo (tag v1.0-demo): https://github.com/v...yak-guardian-ai
Docs: architecture · benchmarks · BOM v2.0 · risks/calibration · ROS2_HYBRID · Pico firmware · deploy services
Showcase PR: https://github.com/D...hallenge/pull/8

**Stage 2 (approved, PR #6):** concept design · AI system architecture · ROS 2 node graph · CPU affinity table · BOM/roadmap/risks · repo structure

**Stage 3 (submitted, PR #8 - review items addressed):** YOLOv8x on the BPU (174 ms / 5.7 FPS) · Whisper + Gemma + Piper offline voice · real ROS 2 nodes on /sahayak/detections · ultrasonic+YOLO sensor fusion · Pico firmware watchdog (0.6 s) + EMI auto-reconnect · Guard/Patrol/Follow/Explore/Check-Room · hands-free wake-word with VAD continuous listening · ESP32 home automation by voice · Gemini action agent ("it's too dark in here" -> light turns on)

Happy to answer anything about the build

[vishal095015]

https://youtu.be/Giq...ceLUnzYw2uhJusT

[vishal095015]

please read this first

[vishal095015]

hi mam, during this competition i know i did so many mistakes and for that sorry, and i will try to fix this all, and i am doing this all not for winning any competition just i want to explore about my highest capability of brain and mind and i love robotics and thank you for giving me this great chance, and i have only one request please again give me this kind of opportunity thank you