AWS DeepRacer: Teaching Machine Learning with AI Race Cars

Why AWS is Building Tiny AI Race Cars to Teach Machine Learning

This article explores the motivations and impact of Amazon Web Services' (AWS) AWS DeepRacer, a 1/18th scale AI race car designed to educate developers about machine learning, particularly reinforcement learning.

The AWS DeepRacer: More Than Just a Toy

The AWS DeepRacer, launched in late 2018, is a small, programmable race car equipped with sensors and software tools that enable developers to build and train machine learning models to navigate a race track. Initially perceived by some as a gimmick, AWS has invested significantly in the DeepRacer platform, fostering a global league where developers can compete and refine their AI models.

Bridging the Gap in Machine Learning Education

Ryan Gavin, AWS's general manager for Artificial Intelligence and Machine Learning marketing, explains that DeepRacer, along with its predecessor DeepLens (a smart camera), stems from AWS's mission to make machine learning accessible to every developer and data scientist. Hardware devices like DeepRacer are seen as crucial for demystifying complex technologies like deep learning and reinforcement learning.

Reinforcement Learning Made Approachable

Reinforcement learning, a machine learning technique that allows agents to learn through trial and error without requiring pre-existing training data, is a key focus. While powerful, reinforcement learning has traditionally presented significant barriers to entry. DeepRacer aims to lower these barriers by providing a tangible and engaging way to understand its principles. Gavin highlights that autonomous vehicles are a relatable application for reinforcement learning, making the concept of training a car to drive itself an intuitive starting point for developers.

The Power of Physical Interaction

Developers who have interacted with DeepRacer at AWS events have reported that the physical nature of the device makes the learning process more enjoyable and accessible. The iterative nature of training a model, where each lap around the track represents a new learning opportunity, resonates with developers. The competitive aspect, with developers pitting their skills against each other globally, further enhances engagement.

Simulation and Real-World Application

AWS leverages cloud-based simulation environments to allow developers to train their models efficiently. This virtual training allows for rapid iteration, simulating thousands of laps per hour. Once a model is sufficiently tuned in the simulation, developers can then deploy it to the physical DeepRacer car for real-world testing.

The Complete Experience: Why AWS Builds the Cars

Instead of merely providing an SDK for developers to build their own cars, AWS chose to offer a complete, working device. Gavin emphasizes that this approach is essential for achieving their goal of making the process approachable. By providing a fully functional car, AWS allows developers to focus on learning reinforcement learning rather than spending time on hardware tinkering.

Driving Adoption and Skill Development

The DeepRacer league has proven successful in driving adoption and skill development. Companies across various sectors, including oil and gas and financial institutions, are hosting internal DeepRacer leagues to upskill their employees in modern machine learning techniques. Educational institutions are also exploring incorporating DeepRacer into their curricula.

Progress and Future Outlook

The progress in DeepRacer competitions has been remarkable, with lap times dramatically decreasing over time, showcasing the rapid learning capabilities of the models. AWS is actively listening to developer feedback, which includes a desire to customize and tinker with the cars. Future developments are expected to be announced at upcoming events like re:Invent, the finals of the racing league.

Key Takeaways:

AWS DeepRacer: A 1/18th scale AI race car for learning machine learning, especially reinforcement learning.
Accessibility: Aims to make complex AI concepts more approachable for developers.
Reinforcement Learning: Focuses on trial-and-error learning without pre-existing training data.
Engagement: Combines physical hardware with competitive leagues for a hands-on learning experience.
Cloud Simulation: Utilizes cloud-based simulations for efficient model training.
Complete Solution: Provides a ready-to-use device to streamline the learning process.
Industry Adoption: Being adopted by companies for employee training and by educational institutions.
Continuous Improvement: Rapid advancements in model performance are evident in competition results.

Topics Covered:

AI & Machine Learning
Cloud Computing
Developer Education
Autonomous Systems
Robotics

Image Credits:

Phillip Faraone/Getty Images for AT&T

Event Promotion:

TechCrunch All Stage pass promotion for Boston event in July 2025.