SelfDrivingCar101
How self-driving cars perceive their surroundings and make driving decisions
tech · autonomous driving · AI · robotics · computer vision · transportation · By 哈哈哈 · 1 learner
What you will learn: How sensors (camera, lidar, radar) perceive the road; sensor fusion basics; localization; perception tasks like detection and tracking; prediction and motion planning; modular pipelines vs end-to-end learning; SAE autonomy levels and current challenges.
Out of scope: Specific algorithm implementations, control theory math, SLAM internals, safety certification, simulation tooling, and code-level deep learning architectures.
Knowledge map
Autonomy Foundations
Frame the self-driving problem and learn the SAE level taxonomy that organizes the field.
Sets up the vocabulary and historical context for the rest of the map. Defines what autonomous driving is, introduces the SAE J3016 levels, and previews the two architectural camps — modular pipelines versus end-to-end learning — that recur throughout later modules.
Sensors for Self Driving
Understand what each sensor type sees, and why a self-driving car uses many of them together.
Covers the eyes and ears of a self-driving car. Establishes the strengths, weaknesses, and operating principles of cameras, lidar, radar, ultrasonic, IMU and GNSS — the vocabulary every later module assumes.
Perception and Sensor Fusion
Understand the perception tasks that turn raw sensor data into a model of the world, and how multi-sensor fusion makes that model robust.
Where raw sensor streams become a structured scene. Covers object detection, semantic segmentation, and multi-object tracking, plus the early, middle and late fusion strategies that combine camera, lidar and radar into one coherent view.
Localization and HD Maps
Understand how a self-driving car knows precisely where it is in the world, and why high-definition maps matter.
Bridges sensing and planning. Explains the localization problem, the role of HD maps as a strong prior on the world, and the GNSS-, lidar- and vision-based techniques used to solve it in practice.
Prediction Planning and Control
Understand how a self-driving car forecasts other agents' behavior, chooses a trajectory, and gets that trajectory onto the road.
The thinking and acting layer. Covers behavior prediction of other road users, the constraints a motion planner must satisfy, how high-level decision making handles intersections and merges, and how a low-level controller turns the chosen trajectory into steering, throttle and brake commands.
Architectures and Open Challenges
Compare modular and end-to-end approaches, and grasp the open problems that keep full autonomy out of reach.
Pulls the whole map together. Revisits the modular versus end-to-end debate with the context built in earlier modules, surveys connected infrastructure such as V2X, and then surveys the long-tail, regulatory and safety challenges that define the frontier of the field.