Risk Estimation with a Learning AI

Activities

RELAI Project

Problem 1

Problem

The likelihood of people walking into the street at the same time on both sides of a shopping street is much higher on a weekday at 3:00 pm than on a Sunday. An attentive, considerate driver knows these relationships, estimates what is currently available and then chooses an appropriate driving style. In the first case, the driver would proactively reduce the vehicle speed to prevent the emergence of dangerous situations or an accident. This dynamic risk assessment is an essential function for future autonomous vehicles whose development and safeguards are still lacking the relevant test scenarios for OEMs and suppliers.

Problem 2

To address this challenge, the project team will take the following steps:

  • Existing data sets that describe critical driving situations are examined for driving context parameters together with suitable data from the mCLOUD and refined accordingly.
  • Through KI-based learning methods, models are developed and trained that generalize these test scenarios. 
  • Thus, new variant-rich synthetic test scenarios can be derived and automatically transferred to different test environments. 
  • For this purpose, a web portal is developed, which is based on the EDI hive standard platform, and is directly connected to the mCLOUD.

Building Blocks

AI-based Test Scenario Generator

AI-based Test Scenario Generator

This algorithm is the core of the RELAI project: by AI-based analysis of test data from road tests and simulations, generic test scenarios are identified and relevant variants of each test scenario are created. In doing so, the consideration of the driving context parameters with focus on the expectations of road users is a key factor to meaningful test scenarios that are useful for the development and validation of autonomous driving functions.

Test Scenario Catalog

Test Scenario Catalog

The generalized, variant-rich synthetic test scenarios of RELAI will be bundled in the Test Scenario Catalog. This database is a central outcome of this project and supports OEMs and suppliers to efficiently develop and validate future autonomous driving functions.

Test Data Analyzer

Test Data Analyzer

The RELAI Test Data Analyzer provides access to the test scenarios of RELAI. Users will be supported by finding relevant scenarios for their specific development tasks (e.g. a lane-keeping assistant requires certain test scopes). Furthermore, users can upload data of simulations or road tests and the application evaluates which scopes of the test scenario catalogue of RELAI have been completely tested and which scenarios have not been considered. 

Automated Labeling of Traffic Scenarios

Automated Labeling of Traffic Scenarios

The aim of the Automated Labeling of RELAI is the detection of road users and identification of their trajectories. The automated consideration of the latest Data Protection Regulations is an essential part of this Building Block.

VR Simulator for Pedestrians

VR Simulator for Pedestrians

The RELAI VR Simulator for Pedestrians captures data about the physical activity of pedestrians using multiple sensors (movement, hand gestures, head movement, point of attention, etc.) in order to transfer it into the simulation and for future behaviour analysis. The specific focus of analysis in the RELAI project is the expectation of the pedestrian in different scenarios.

Road User Expectation Model

Road User Expectation Model

Based on the studies with pedestrians conducted during RELAI and additional simulation and road test data a model that describes the expectations of road users in different scenarios will be generated. This model is used to evaluate the safety of behaviour of vehicles and road users in traffic depending on the driving context.