Work

AWS DeepRacer - Reinforcement Learning Competition

Machine Learning
Python
AWS
Reinforcement Learning

Developed and optimized a reinforcement learning model for autonomous racing competition, achieving top 0.25% performance among 5,200+ global submissions.

AWS DeepRacer autonomous racing car on a track

Project Overview

AWS DeepRacer is a 1/18th scale autonomous racing car platform that provides a practical way to learn reinforcement learning. The challenge was to develop a model that could navigate complex racing tracks at relatively high speeds, achieving competitive lap times.

Technical Implementation

Reward Function Engineering
  • Designed a sophisticated reward function incorporating multiple parameters:
    • Track position and centerline deviation
    • Steering angle and stability metrics
    • Speed optimization based on track segments
    • Progress tracking and completion incentives
  • Implemented dynamic rewards scaling based on racing conditions
Hyperparameter Optimization
  • Fine-tuned critical parameters including:
    • Learning rate: Optimized for convergence speed
    • Entropy and discount factor
    • Neural network architecture
    • Action space granularity
  • Developed all-new method to assess model performance
Training Infrastructure
  • Using AWS SageMaker for distributed training
  • Implemented parallel training pipelines for faster iteration
  • Created automated evaluation scripts for performance tracking
  • Optimized compute resource utilization to reduce costs

Results & Impact

  • Achieved ranking in top 0.25% globally (13th out of 5,200+ submissions), receiving recognition from AWS
  • Reduced training costs by 40% through efficient resource management (first vs. final model)

Technical Stack

  • AWS Services:

    • DeepRacer Console
    • SageMaker
    • RoboMaker
  • Languages & Frameworks:

    • Python
    • TensorFlow
    • Pandas for data analysis

Key Learnings

  • Deep understanding of reinforcement learning principles
  • Practical experience with hyperparameter tuning
  • Efficient cloud resource management
  • Algorithm optimization, for smart decision-making