Simulation in Automotive: Training and Validating Autonomous Control Systems
Waymo, Uber, Ford and, BMW are among the multiple players who have committed to bring SAE level 4 or level 5 vehicles to market within the early years of the next decade. Almost all OEMs have positioned connected, automated vehicles as the cornerstone of their future strategy, hailing the advantages they will bring in terms of safety and efficiency.
However, with the self-imposed deadlines for implementation fast-approaching, OEMs are still struggling to validate their autonomous vehicles and have confidence that they will be significantly safer than manually-controlled vehicles. Far from being a question of better sensors or greater processing power, the biggest barrier to deployment is now experience – ramping up the volume and variety of the situations which autonomous systems have navigated. With OEMs needing billions of miles of experience to have confidence that their autonomous systems are significantly safer than their manually-driven equivalents, the status quo of small fleets of autonomous prototypes is clearly unsustainable.
Therefore, a number of autonomous system developers are embracing simulation as a means to artificially generate the volume and variety of autonomously driven miles needed to validate their solutions. While Waymo has already effectively leveraged simulation effectively through their in-house Carcraft initiative, a number of start-ups and new product launches are helping OEMs to close the gap.
This report analyses the relationship between simulation and deep learning approaches in the automotive industry, the key advantages and use-cases, and sets out the core technologies and competencies that OEMs should look for when choosing their simulation partners.
Table of Contents
- 1. EXECUTIVE SUMMARY
- 1.1. Billions of Miles of Experience
- 1.2. Sensor Entrenchment
- 1.3. Traceability and Type Approval
- 2. INTRODUCTION AND KEY TERMS
- 3. KEY APPLICATIONS FOR SIMULATION IN AUTOMOTIVE
- 3.1. Self Certification and System Validation
- 3.2. Testing Corner Cases
- 3.3. Testing To Destruction
- 3.4. Public Agency Type Approval and Testing
- 3.5. Sensor Evaluation and Competitive Benchmarking
- 4. CORE TECHNOLOGIES AND COMPETENCIES FOR SIMULATION
- 4.1. Environment Modeling
- 4.2. AI Agents
- 4.3. Scenario Creation
- 4.4. Sensor Emulation
- 4.5. Autonomous Control System: HIL and SIL Testing
- 4.6. Feedback into Environment Modeling
- 4.7. 3D Visualization
- 5. ECOSYSTEM
- 5.1. Waymo Carcraft
- 5.2. NVIDIA Constellation
- 5.3. Cognata
- 5.4. Siemens Simcenter
- 5.5. Metamoto