Problem Statement

Scope an autonomous-driving perception problem to semantic image segmentation. Compare the information supplied by different sensors, then distinguish image classification, object detection, semantic segmentation, and instance segmentation. These outputs support downstream decisions; segmentation alone does not define a complete driving system.

Introduction

The definition of a self-driving car is a vehicle that drives itself, with little or no human intervention. Its system uses several sensory receptors to perceive the environment. For instance, it identifies the drivable area, weather conditions, obstacles ahead and plans the next move for the vehicle accordingly.

Distinguish driver assistance from an automated driving system operating within defined conditions. Tesla’s documentation states that Full Self-Driving (Supervised) requires driver supervision and does not make the vehicle autonomous. Waymo’s safety-methodology paper describes a Level 4 system and assessment within an operational design domain, not unrestricted driving under every condition. Goods-delivery vehicles, such as the Nuro example in the source, pose related perception tasks.