How Autonomous Cars Process Sensor Data for Driving

Intel's 2020 patent describes a system for autonomous vehicles that cleans and standardizes data from various sensors before using it to perceive the environment and make driving decisions.

What it covers

This patent details an apparatus, likely a computer system within an autonomous vehicle, designed to handle the raw information gathered by the car's sensors, like cameras and lidar. The system first receives this 'sensor data' through an interface. Then, its 'processing circuitry' (the car's computer brain) applies an 'abstraction process' to this data. This process involves cleaning up the data by normalizing its values, correcting distortions through warping, or removing noise via filtering. The goal is to create 'abstracted scene data' that is reliable and consistent. This cleaned-up data is then used in the 'perception phase' of the car's control system, helping it understand what's around it, like other cars, pedestrians, and road signs, to make safe driving decisions.

What it doesn't cover

• —Does not cover the specific hardware components of the sensors themselves.
• —Does not cover the final decision-making or actuation (steering, braking) of the autonomous vehicle, only the data processing leading up to it.
• —Does not cover systems that do not abstract sensor data before the perception phase.
• —Does not cover methods of abstracting sensor data that do not involve normalization, warping, or filtering.
• —Does not cover sensor data from non-autonomous vehicles.

The clever bit

The innovation lies in creating a standardized 'sensor abstraction process' that can handle data from different types of sensors (like cameras and lidar) and even multiple sensors of the same type. This abstraction layer ensures that the perception system receives consistent, high-quality data, regardless of the original sensor's quirks or environmental conditions.

Why it matters

This patent is significant because it addresses a fundamental challenge in autonomous driving: making sense of diverse and potentially noisy data from multiple sensors. Standardizing and cleaning this data is crucial for reliable perception, which underpins the safety and functionality of self-driving cars. It represents Intel's effort to provide foundational technology for the burgeoning autonomous vehicle industry.

Real-world examples

• 1.Autonomous vehicle sensor data processing systems
• 2.Advanced Driver-Assistance Systems (ADAS)
• 3.Self-driving car perception modules