Want to Make your Every Day Safer? You Can Start by Using More Accurate Addresses

By Ashley Marie Cashion, VP USA & Head of...

Want to Make your Every Day Safer?...

Discovering the 'Unknown' Unknowns in Autonomous Vehicle Testing Through Photorealistic Simulation

By Danny Shapiro, Senior Director of Automotive,...

Discovering the 'Unknown' Unknowns in...

Functional Safety takes the Driver's Seat in Autonomous Vehicles

By Neil Stroud, Senior Director of Technology...

Functional Safety takes the Driver's...

People, Not Technology, Stand in the Way of an Autonomous Future

By Alexei Andreev, Managing Director, Autotech...

People, Not Technology, Stand in the...

How the Advancement of Autonomous Vehicles Relies on Data Acquisition

By Dr. Florian Baumann, CTO (Specializing in Automotive & AI), Dell EMC

How the Advancement of Autonomous Vehicles Relies on Data AcquisitionDr. Florian Baumann, CTO (Specializing in Automotive & AI), Dell EMC

As advancements develop within the autonomous driving industry, the requirements for advanced driver-assistance systems (ADAS) have risen significantly.

An autonomous-connected vehicle requires a variety of safety features in order to gather enough information about the surrounding traffic environment. It needs a multitude of cameras, laser scanners, radar sensors, and ultrasonic sensors. It also needs a differential global positioning system (GPS) and an inertial measurement unit (IMU) to gauge vehicle dynamics and movements.

The sensor data is becoming more complex and connected vehicles must be able to operate in all countries of the world under a range of conditions to safely and successfully drive autonomously. This requires an inordinate amount of data, yet algorithms and methods from the field of machine learning are reaching capacity.

This exorbitant amount of data is processed in real-time from the vehicle’s electronic control unit (ECU) to enable the car to drive autonomously and safely. But during the development phase of the algorithms that process the data, the data must be stored within the vehicle to train, test, and validate the machine learning algorithms.

Autonomously driving vehicles are equipped with up to 30 different sensors producing up to 10 terabytes (TB) of data per hour. To realize a robust sensor fusion between the different types of sensors in connected vehicles, two critical actions are necessary during the recording: calibration and synchronization.

To realize a robust sensor fusion between the different types of sensors in connected vehicles, two critical actions are necessary during the recording: calibration, and synchronization

All sensors must calibrate extrinsically to each other, which means the position and rotation between sensor and real-world must be known. Furthermore, it is crucial that the sensors synchronize with each other with accuracies of up to 100ns (nanoseconds). In other words, a global timestamp is assigned to every image frame or incoming sensor message.

During the data acquisition phase of ADAS and autonomous driving system development, cars are continuously driving in the real world to collect sensor data under different lighting and environmental conditions, in different countries with differing road systems, under all weather conditions and on all types of roads.

Every development car has a data acquisition system that receives data from all kinds of sensors and stores them synchronized in a specific data format on the in-car storage system. Write and read capabilities of up to 100 gigabits per second are needed. Since the cars are collecting data around the clock, the in-car storage unit often relies on removable storage cartridges. Thus, drivers can easily remove cartridges replete with storage and replace them with empty ones to send the car immediately back to the recording campaign.

Typical data formats to store sensor data are based on Robot Operating System (ROS) bag files, which are open-source and typically used in the development of autonomous driving systems. An automated data and time-triggered framework system (ADTF) is often used by companies focusing on ADAS in the production.

The last step is to ingest the recorded sensor data to the central data center, either virtually or physically. The following options are chosen by automotive companies according to their requirements:

• Shipping storages cartridges through postal services and using custom-ingest stations.

• Using a mobile upload station, capable of storing petabytes of data.

• Directly connecting the car through cabled connection to the data center.

• Transferring data wirelessly through transfer pods.

• Sending data over via 4G or 5G networks.

Different data acquisition and ingestion solutions are on the market. It is imperative to focus also on ingestion of data and to follow the complete workflow until the data is located in the data lake and ready for the developers to train the algorithms.

Dell Technology, the world’s leading enterprise IT Infrastructure Company, is delivering complete End-to-End solutions for the ADAS and AD product development, which will have a direct impact on reducing the operational risk, complexity, and accelerating the time to value.

The long experience in the automotive industry supporting various customers, enables Dell Technologies to proactively support the evolution of product development of AD/ADAS environments.

Read Also

Wearable Robotics - Opportunity for the Automotive Industry

Wearable Robotics - Opportunity for the Automotive Industry

Sigune Suttner, Founder and CEO, StraightWalk
Want to Make your Every Day Safer? You Can Start by Using More Accurate Addresses

Want to Make your Every Day Safer? You Can Start by Using More Accurate Addresses

Ashley Marie Cashion, VP USA & Head of Automotive & Mobility, what3words
Discovering the 'Unknown' Unknowns in Autonomous Vehicle Testing Through Photorealistic Simulation

Discovering the 'Unknown' Unknowns in Autonomous Vehicle Testing Through Photorealistic Simulation

Danny Shapiro, Senior Director of Automotive, NVIDIA (NASDAQ: NVDA)
Functional Safety takes the Driver's Seat in Autonomous Vehicles

Functional Safety takes the Driver's Seat in Autonomous Vehicles

Neil Stroud, Senior Director of Technology Strategy, Automotive & IoT Line of Business, Arm
People, Not Technology, Stand in the Way of an Autonomous Future

People, Not Technology, Stand in the Way of an Autonomous Future

Alexei Andreev, Managing Director, Autotech Ventures
How the Advancement of Autonomous Vehicles Relies on Data Acquisition

How the Advancement of Autonomous Vehicles Relies on Data Acquisition

Dr. Florian Baumann, CTO (Specializing in Automotive & AI), Dell EMC
follow on linkedin follow on twitter Copyright © 2020 www.autotechoutlook.com All Rights Reserved | Privacy Policy
Top