FailSim AI ⚡

Multi-Robot Failure Discovery & Root-Cause Analysis Platform

---

🚨 The Problem

Robotics companies generate massive simulation data but struggle to understand:

• Which combinations cause failures?
• What are the root physical causes?
• How to prioritize fixes?

Traditional approach:

• ❌ Manual log review by engineers
• ❌ Slow, expensive, error-prone
• ❌ Misses subtle patterns

FailSim AI approach:

• ✅ Automated physics-based testing
• ✅ AI-powered root cause analysis with Gemini 3 Flash
• ✅ Multi-robot support (Kuka, Franka Panda, UR5, + Custom)
• ✅ Real-time visualization & charts

---

💡 Our Solution

FailSim AI is a physics-based simulation platform that automatically discovers failure patterns across multiple robot types and uses Gemini 3 Flash AI to explain WHY robots fail.

Real-World Use Case

Company: Warehouse automation startup
Robot: Kuka iiwa7 collaborative arm
Challenge: Pick & place fails unpredictably in production

Before FailSim AI:

"Our robot fails sometimes. Let's manually test 100 combinations and guess why."
Cost: 2 weeks, $10K in engineer time

With FailSim AI:

"Run 1000 simulations overnight. FailSim AI reports: 'Failures occur when required grip force exceeds 100N (weight > 1.1kg AND friction < 0.32). Recommendation: Increase gripper force by 15%'"
Cost: 1 day, automated

---

🤖 What FailSim AI Does

1. Multi-Robot Physics Simulation

• 3 Built-in Robots:
  • Kuka iiwa7 (100N force, 7kg payload)
  • Franka Emika Panda (70N force, 3kg payload)
  • Universal Robots UR5 (150N force, 5kg payload)
• Custom Robot Support: Users can define their own robot specs
• Real Physics: PyBullet 3D engine with actual equations
• Domain Randomization: Weight, friction, lighting variance

2. Automatic Failure Discovery

• Runs 100s of simulations
• Physics-driven failures (not random):
  • Gripper force insufficient
  • Object slips during acceleration
  • Vision errors from poor lighting

3. AI-Powered Analysis (Gemini 3 Flash)

• Analyzes failure clusters
• Generates human-readable explanations
• Provides engineering recommendations
• Fallback to Gemini 2.5 Flash if quota exceeded

4. Interactive Features

• Experiment Runner: Configure and run simulations from dashboard
• Live Visualization: See 5 key frames of robot simulation
• Click-to-Analyze: Get AI analysis for individual runs
• Failure Charts: Weight/friction distribution visualizations
• Custom Robots: Add your own robot specifications

---

🏗️ Architecture

┌─────────────────┐
│   PyBullet      │  Multi-Robot Physics Simulation
│   3 Robots      │  • Kuka iiwa7, Franka, UR5
│   + Custom      │  • Real physics equations
└────────┬────────┘
         │ Simulation Results (JSON)
         ▼
┌─────────────────┐
│  Vultr Backend  │  Central System of Record
│  FastAPI        │  • REST API (Port 8000)
│  Frankfurt VM   │  • Experiment orchestration
└────────┬────────┘
         │ Failure Data
         ▼
┌─────────────────┐
│  Gemini 3 Flash │  AI Analysis
│  (+ 2.5 fallback)│  • Root cause analysis
│                 │  • Physics explanations
└────────┬────────┘
         │ AI Insights
         ▼
┌─────────────────┐
│  React          │  Interactive Dashboard
│  Dashboard      │  • Multi-robot selection
│  Port 80        │  • Live visualizations
└─────────────────┘

Tech Stack:

• Simulation: PyBullet, Python 3.10, NumPy
• Backend: FastAPI, Vultr VM (Frankfurt)
• AI: Google Gemini 3 Flash API (with 2.5 fallback)
• Frontend: React, Tailwind CSS, Vite
• Deployment: Nginx, Screen sessions

---

🚀 Quick Start

Prerequisites

• Python 3.10+
• Node.js 18+
• pip, npm

Local Development

# Clone repository
git clone https://github.com/SanaAdeelKhan/FailSim-AI.git
cd FailSim-AI

# Setup Python environment
python3 -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
cd backend
pip install -r requirements.txt

# Add Gemini API key
echo "GEMINI_API_KEY=your_key_here" > .env

# Run backend
python src/main.py
# Backend: http://localhost:8000

# New terminal: Run frontend
cd ../frontend
npm install
npm run dev
# Dashboard: http://localhost:5173

---

📊 Live Demo

🌐 Dashboard: http://80.240.20.49
🔌 API Docs: http://80.240.20.49:8000/docs

Try it:

# Get stats
curl http://80.240.20.49:8000/api/stats

# Get AI insights
curl http://80.240.20.49:8000/api/insights

# Get available robots
curl http://80.240.20.49:8000/api/robots

---

🎯 Key Features

✅ Multi-Robot Support

• 3 pre-configured industrial robots
• Custom robot creation with user-defined specs
• Different physics parameters per robot type

✅ Real Physics

• Actual robot specifications (gripper force, payload, acceleration)
• Physics equations: F_grip = (m × g + m × a) / μ
• No arbitrary probabilities

✅ Gemini 3 AI Integration

• Automatic fallback to Gemini 2.5 if quota exceeded
• Failure pattern analysis
• Physical mechanism explanations
• Actionable recommendations

✅ Interactive Dashboard

• Experiment Runner: Configure simulations (runs, robot type)
• Live Visualization: 5-frame robot simulation capture
• Click-to-Analyze: Individual run AI analysis
• Charts: Error distribution, weight/friction patterns
• Custom Robots: Add your own robot specs on-the-fly

✅ Production-Ready

• REST API for integration
• Scalable batch processing
• 24/7 deployment on Vultr
• Web dashboard with real-time updates

---

👥 Team

Built by Team FailSim for Launch & Fund Hackathon:

• Sana Adeel - AI Architecture & Full-Stack Development
• Wajiha Saleem - Simulation & Physics Engineering
• Tooba Muzaffar - Frontend & UX Design
• Ghulam Hussain Ali - Backend Infrastructure
• Muhammad Zargham Khan - Documentation & Testing

---

🎯 Target Users

• Robotics Startups: Reduce hardware testing costs
• Autonomous Systems Teams: Discover edge cases early
• QA Engineers: Systematic failure analysis
• Research Labs: Validate before real-world deployment

---

💼 Business Value

ROI Example:

Approach	Time	Cost	Scenarios
Traditional QA	2 weeks	$10K	50
FailSim AI	1 day	~$0	1000+
Result	10x faster	90% cheaper	20x coverage

---

🏆 Hackathon

Event: Launch & Fund - AI Meets Robotics
Track: Track 2 - Simulation-to-Real Training & Evaluation
Dates: February 6-14, 2026
Organizer: lablab.ai

Compliance Checklist

• ✅ Track 2: Simulation-to-Real evaluation pipeline
• ✅ Vultr VM backend (mandatory)
• ✅ Public web application
• ✅ Gemini AI integration (3 Flash + 2.5 fallback)
• ✅ Software-only, simulation-first
• ✅ Production-ready web app
• ✅ GitHub repository with docs

---

📄 License

MIT License - see LICENSE file

---

🙏 Acknowledgments

• Vultr - Cloud infrastructure partner
• Google - Gemini AI API access
• PyBullet - Physics simulation engine
• lablab.ai - Hackathon platform

---

FailSim AI — Because finding failures early is the fastest way to ship reliable robots. 🚀

Built with ❤️ by Team FailSim