BMW launches D3 platform for data-driven development of autonomous driving
The launch of the new BMW Group High Performance D3 platform represents a key milestone on the BMW Group’s roadmap to highly and fully automated driving.
The “D3” in the new IT platform's name stands for Data-Driven Development, which forms the basis for the development and validation of highly and fully automated driving functions. Data-Driven Development is an indispensable tool in securing the safety and reliability of the Level 3 system to be offered in the BMW iNext in late 2021.
The BMW Group has been applying the Data-Driven Development approach for a couple of years now. The basic principle is rooted in the assumption that the only way of mapping - and thereby ultimately handling - the complexity and variety of traffic situations encountered on every continent is to gather vast quantities of data. This means the algorithms and overall operation of autonomous driving have to be validated using a broad data pool.
The first step in the process is to collect approx. 5 million kilometres (3.1 - 3.7 million miles) of real-life driving data from the test fleet vehicles. From this data, two million kilometres (1.25 million miles) of the most relevant driving scenarios and environmental factors are then extracted.
The relevance of the data collected is continuously improving thanks to the way in which qualitative data is selected using data qualification/filtering. These two million kilometres of driving data subsequently undergoes regular reprocessing as development progresses. This happens whenever a new control unit integration level (I-level) becomes available, in order to evaluate the new I‑level's increase in performance.
This qualified two million kilometres of data is constantly expanded by a further 240 million kilometres (150 million miles) of simulation-generated data, which is primarily based on the relevant driving scenarios and ensures that the immense diversity of real-life driving is taken into account properly during development.
The reprocessing of the two million real-life kilometres and 240 million virtual kilometres requires a high performance data platform of over 230 petabytes storage capacity and the computing power of more than 100,000 cores and more than 200 GPUs (Graphics Processing Units).
There is a 96 x 100Gbps connection between the BMW Group High Performance D3platform and the Hardware-in-the-Loop (HiL) stations located at the BMW Group Autonomous Driving Campus. The net useable data rate is approx. 3.75 Terabit/s.
The fleet currently numbers around 80 BMW 7 Series cars, which are in operation on the west coast of the US, in Germany, Israel and China. The number of vehicles is set to increase to approx. 140 by the end of 2019.
Below are some figures to illustrate the capabilities of the BMW Group High Performance D3 platform:
1. Daily collection of more than 1.500 TB raw data
2. D3 Platform storage capacity of more than 230 PB
3. Compute: > 100.000 Cores and > 200 GPU´s
4. 50 PB to HiLs every two weeks
For the BMW Group High Performance D3 platform, the German automaker has partnered with DXC Technology. The core component of DXC's work is to setup and run the data centre and to develop applications with the objective to support the autonomous driving development process. The aim is to reduce costs and the time needed until the system is ready to market.
The digital solutions from DXC put the BMW Group's development teams in a position to collect, store and manage the data from vehicle sensors - and make it available for the requisite AI training - in a matter of seconds rather than days or even weeks.
The DXC solution was developed in an open source environment and is available on-premise and in a hybrid environment, allowing workloads to be shifted easily. This paves the way for agile cooperation between engineers, regardless of their location.
Using a single platform for data storage, processing and AI training lowers the hardware and software requirements, thereby reducing costs and complexity. Data can be gathered globally but monitored centrally. This has the effect of maximising efficiency and cutting costs.
The BMW Group High Performance D3 platform was completed in the space of a few months, in function, on time and on budget. It is sited just a few kilometres away from the BMW Group Autonomous Driving Campus in Unterschleißheim near Munich. The close proximity was an essential part of the plan to transfer the enormous quantities of data from the Campus to the platform via cable.
BMW launches D3 platform for data-driven development of autonomous driving
Modified on Thursday 28th March 2019
Find all articles related to:
BMW launches D3 platform for data-driven development of autonomous driving