🤖 Robot Navigation Challenge

A technical assessment for AI + Robotics engineering positions. Train a neural network to control a simulated differential-drive robot using imitation learning.

📋 Overview

In this challenge, you'll work with a codebase that trains a neural network to navigate a two-wheeled robot to target positions. The code has issues that cause poor real-world performance — your task is to investigate, diagnose, and fix them.

Time Estimate: 3.5 - 4.5 hours

🚀 Quick Start

Prerequisites

• Python 3.11 or higher
• pip package manager
• Git

Setup

# Clone the repository
git clone https://github.com/gabo-di/RoboAI.git
cd RoboAI

# Create and activate virtual environment
conda create --name roboai python=3.11
conda activate roboai

# Install dependencies
pip install -r requirements.txt

Run the Pipeline

# Step 1: Train the model
python src/train.py

# Step 2: Evaluate performance
python src/evaluate.py

📁 Project Structure

robot-nav-challenge/
├── src/                             # Source code
│   ├── robot_simulator.py           # Robot physics simulation
│   ├── dataset.py                   # Data generation & splitting
│   ├── model.py                     # Neural network architectures
│   ├── utis.py                      # Some utils
│   ├── train.py                     # Training script
│   └── evaluate.py                  # Evaluation script
├── tests/                           # Unit tests
├── configs/                         # Configuration files
│   ├── dataset_default.toml         # Dataset default configs
│   ├── evaluate_default.toml        # Evaluation default configs
│   ├── robo_simulator_default.toml  # Robo simulator default configs
│   └── train_default.toml           # Train default configs
├── docs/                            # Documentation
│   ├── PROBLEM_STATEMENT.md
│   ├── SOLUTION_REPORT_TEMPLATE.md
│   └── SIDE_QUESTS.md
├── requirements.txt
├── LICENSE
├── CONTRIBUTING.md
├── requirements.txt
└── README.md

📖 Challenge Instructions

Read the full problem statement: docs/PROBLEM_STATEMENT.md

Optional bonus challenges: docs/SIDE_QUESTS.md

🔧 For Candidates

Submission Guidelines

1. Fork this repository to your own GitHub account
2. Create a feature branch for your work:

   git checkout -b solution/firstname-lastname

3. Make your changes with clear, atomic commits
4. Document your findings in docs/SOLUTION_REPORT.md
5. Push your branch and create a Pull Request

Commit Message Convention

We follow Conventional Commits:

feat: add sin/cos angle preprocessing
fix: correct train/val data leakage
docs: add solution report
test: add unit tests for robot simulator
refactor: extract motor calibration to config

What We're Looking For

• Systematic debugging approach
• Clear understanding of ML and robotics concepts
• Quality of implemented solutions
• Code quality and documentation
• Bonus: Creative solutions that surprise us!

📊 Expected Output

After training, you should see:

• Training curves saved to training_curves.png
• Model checkpoint in checkpoints/best_model.pt
• Evaluation results comparing ideal vs realistic robot performance

🐳 Docker (Optional Bonus)

See docs/SIDE_QUESTS.md for bonus challenges including Docker containerization.

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

❓ Questions?

If something is unclear about the challenge requirements, please open an issue with the question label.

---

Good luck! 🍀