Case Studies > Put a Label on It

Put a Label on It

Company Size

1,000+

Region

Europe

Country

Germany

Product

understand.ai
dSPACE

Tech Stack

Neural Networks
GPU-based Computer Clusters
3-D Annotation Tools

Implementation Scale

Enterprise-wide Deployment

Impact Metrics

Customer Satisfaction
Innovation Output
Productivity Improvements

Technology Category

Analytics & Modeling - Machine Learning
Analytics & Modeling - Predictive Analytics
Application Infrastructure & Middleware - Data Visualization

Applicable Industries

Automotive
Software

Applicable Functions

Product Research & Development
Quality Assurance

Use Cases

Machine Condition Monitoring
Predictive Maintenance

Services

Data Science Services
System Integration
Training

About The Customer

Bosch is a leading global supplier of technology and services, operating in various sectors including automotive, industrial technology, consumer goods, and energy and building technology. With a strong focus on innovation and quality, Bosch aims to enhance the quality of life with solutions that are both innovative and beneficial. The company employs over 400,000 associates worldwide and is known for its pioneering work in the field of autonomous driving. Bosch collaborates with various partners and leverages cutting-edge technologies to develop reliable and efficient systems for autonomous vehicles. Their commitment to research and development is evident in their continuous efforts to explore new technologies and methodologies to improve safety and efficiency in mobility solutions.

The Challenge

New innovative concepts for mobility, such as highly automated or autonomous driving, place enormous demands on the safety and reliability of technical systems. The efficient development of particularly reliable control systems for autonomous driving up to SAE Level 5 requires the use of suitable technologies. Conventional control-based approaches compete with the capabilities of trained neural networks, which are particularly suitable for processing vast amounts of data from high-resolution sensors. The first step is to identify potential application areas for AI along the entire processing chain from perception to situation analysis and behavior planning. Promising methods from machine learning must be evaluated, focusing on multimodal perception, i.e., the environment perception of a vehicle with merged data from video, radar, and lidar sensors.

The Solution

The raw data from sensors installed in vehicles, recorded during real drives, is used as training data. The traffic environments for the recordings are diverse, covering highways, country roads, urban areas, and various traffic scenarios. This diversity is achieved by defining ideal types and characteristics of routes and selecting tracks that match these definitions. Neural networks learn by means of positive and negative examples, a process called supervised learning. The tasks and solutions, typically raw sensor data and detected objects, are required to identify correct results. The solutions are made available by markings (labels/annotations) in the form of reference data. The anonymized data is transferred to understand.ai, a dSPACE group company specializing in labeling automation. Bosch and understand.ai agreed on precise quality goals to ensure successful training. Annotation was performed with data from lidar sensors, marked with high-precision bounding boxes in the 3-D point clouds. Camera sensor data was used for plausibility checks. An iterative approach with feedback loops ensured the required quality level early and continuously.

Operational Impact

Highly accurate annotation is crucial for supervised learning, as the quality of reference data determines the AI's ability to identify objects clearly.
The iterative approach with feedback loops and continuous requirement refinement helped achieve the defined quality goals early and continuously.
The experience and expertise of recognized annotation specialists were invaluable for an economical and efficient approach.
Established processes, powerful tools, and efficient feedback cycles led to the desired results.
Annotation poses special challenges, such as differentiating between cars and vans or detecting vehicles with roof boxes or bicycle racks, which were successfully addressed by understand.ai's web- and AI-based tools.

Quantitative Benefit

Training neural networks can take several days or weeks, depending on the depth of the network and the amount of data.
The project involved recording data from diverse traffic environments, including highways, country roads, and urban areas.