MLOps Team Introduction

TIER IV MLOps Team

ML Team

Table of Contents 01 About MLOps Team 02 MLOps Overview 03 MLOps Services 04 Appendix

About MLOps Team 01

TIER IV Get the System Smarter with Every Mile Strive to create the foundation of an autonomous driving system that gets smarter with every mile. Build an environment where autonomous driving developers can dedicate themselves to enhancing the performance of machine learning models.

TIER IV About Us Tech Blog Presentation MLOps Team Our People

TIER IV Team Objectives Develop and maintain a cloud-based learning pipeline with robust experiment tracking and reproducibility. Automate annotation and extract important training data. Implement efficient distribution of data models to vehicles while maintaining continuous monitoring.

TIER IV Culture Innovative: We leverage web technologies to build necessary tools and pipelines for machine learning development. Proactive: We embody the vision essential for the foundation of autonomous driving that gets smarter with every mile. Communication-focused: We are continuously attuned to the needs of developers engaging in dialogue, and crafting effective features collaboratively.

MLOps Overview 02

TIER IV Web.Auto Overview CI/CD Pipeline A cloud-based infrastructure and editing tools to support the build of autonomous driving software and the execution of extensive test cases through simulation, including scenarios and map editing. Simulation Build Test Dataset, ML Models Development The regeneration of realistic events from autonomous driving logs, scenario-based simulation, and simulation using virtual sensors. Maps, Scenarios Autonomous Driving Dev. Cycle Vehicle Data Data Collection Deployment Firmware Images Operation Data Management Efficient data collection from autonomous vehicles, and data retrieval for learning and testing purposes. Fleet Management Remote Operation Remote driving and monitoring of vehicle status. Management and scheduling of autonomous vehicles, data analysis, and OTA updates.

TIER IV MLOps Workflow Data Collection Data Upload Tool Data Processing Annotation Data Processing Annotation Training Evaluation Training Script Development Deployment Autonomous Driving Software Development 3rd Party Annotation Tool OTA Data Prep Build Raw Rosbag Dataset for Annotation Training Build Simulation Annotated Dataset ML Model Data Service ML Service MLOps Team CI/CD Service

TIER IV Data/ML Components Operator FMS Data/ML Services Data Collection Data Platform Data Processing Vehicle Telemetry Data Log File (rosbag etc) Annotation Integration BI Tool Mobility Observer t4dataset CI/CD Pipeline ML Model Training Pipeline Annotation Tool Data/ML User (Autonomous Driving Developer) Evaluator Scope of Development Control lines Out of Scope Data lines

TIER IV Data/ML Workflow & Teams Data Collection Data Processing Data Labeling Model Development Data Engineers Model Deployment Model Monitoring MLOps Engineers Data/ML Team FMS/Drive Team System Integration Team Sensing/Perception Team MLOps Team CI/CD Team

MLOps Services 03

TIER IV MLOps Services Realize “The more you drive, the more you learn” concept in Autoware. Model Development Model Monitoring Ensure training reproducibility & traceability ML training/evaluation pipeline Model Deployment Ensure model performance on Autoware Model Deployment OTA updates Model Monitoring Evaluate model performance in operation Detect and extract edge cases MLOps Team Model Development

TIER IV Training Pipeline/Model Management Manage Model Releases & CI/CD Pipeline Integration ML Service Trainig Script Build Data Prep Annotated Dataset Training CI/CD Service ML Model Check Training Logs in Real Time List of Training Jobs with Status & Metrics MLOps Team

TIER IV Training Pipeline System Architecture .webauto-ml.yml Source Code Upload Source Code Bucket Web.Auto ML Dashboard Web.Auto CLI Experiment Run Workflow Build Workflow Preprocessing Workflow Training Workflow CodeBuild SageMaker SageMaker ML Service API Server SQS Artifact Bucket Data Management Service DynamoDB ECR Dataset Bucket Annotated Dataset MLOps Team

TIER IV Technology Stack Back end Go Infrastructure Front end Python TypeScript CI/CD React Logging/Monitoring Datadog MLOps Team Sentry GitHub Actions

TIER IV Contact us https://tier4.jp/careers/ Thank you

• https://tier4.jp/careers/

Appendix 04

TIER IV Autoware Open-source autonomous driving operating system Autoware is designed to run on a wide variety of hardware platforms. There is no other autonomous driving software that can be applied to as many types of ECU, sensors and vehicles as Autoware. We offer various services such as integration, risk management and fleet operation. MLOps Team

TIER IV ML Models in Perception Module Dynamic Object 2D Object Detection Dynamic Object Detection Tracking LiDAR based CenterPoint Camera based YOLOX Perception Prediction Traffic Light Traffic Light Detection 3D Object Detection Fine Detection Classifier Classification MobileNet V2 YOLOX MLOps Team

TIER IV Evaluation on Simulator Overview Logging Simulator Sensing ✓✓ Localization ✓✓ Perception ✓✓ Planning Simulator ✓ Sensor Simulator End2End Simulator Real Test ✓✓ ✓✓ ✓✓ ✓✓ ✓✓ ✓✓ ✓✓ ✓ Planning ✓✓ ✓✓ ✓✓ Control ✓ ✓✓ ✓✓ MLOps Team

TIER IV ML Model Deployment Issue Safety All ML models have passed the prescribed test cases and are approved for release No runs are allowed without passing tests Flexibility Use of different ML model for each driving environment and vehicle Firmware Firmware Firmware Firmware Autoware Autoware Autoware Autoware ML Model A ML Model B Vehicle A Vehicle B ML Model A ML Model B ML Model A ML Model B Vehicle A Vehicle B Vehicle A Vehicle B Pros Flexible delivery Pros Flexibility of switching Pros Safety assurance Cons Complexity of pattern management Cons Assurance of switching conditions Cons Switching costs MLOps Team

TIER IV Data Search Console Rosbag Management Annotation Dataset Management MLOps Team