About SIMUSAFE

PROJECT VISION

Road transport is known to be the most dangerous of all transport modes and poses a major societal challenge for the EU. According to the European Commission (ec.europa.eu), road crashes cause almost 30,000 fatalities and more than 100,000 serious injuries a year in the EU. In 2015, there were more than 26,000 road fatalities in the EU:

  • 22% pedestrians
  • 8% bicycle riders
  • 15% motorcycle riders
  • 3% moped (and similar powered two-wheelers) riders
  • 52% motor vehicle occupants

Road crashes also cause tens of thousands of injuries annually ranging from crippling/devastating to minor. It is estimated that associated costs are at least €100 billion a year.[1]

It has been claimed that 90% of road-traffic crashes are caused by driver error with risky behaviour being a significant factor in traffic collisions. Improving road safety means understanding the individual and collective behaviour of all the actors involved (drivers, two-wheelers, pedestrians) and the interactions between themselves, the transport-related systems, and the infrastructure. Therefore, SIMUSAFE will focus on the most at-risk transportation situations by looking at dangerous road designs as well as the altered driving conditions that frequently impair road users.

[1]

PROCESS

In order to build more realistic driving simulators and simulation models, SIMUSAFE will collect and integrate multiple sources of road user data in three research cycles:

  • First, project partners will collect naturalistic driving, riding, and walking behaviours in uncontrolled environments for a baseline.
  • Second, we will collect behavioural and physiological responses under more controlled conditions to connect risk taking behaviour and cognition.

The above data-collection phases will be refined, correlated, and used to create more realistic multi-actor simulation models.

  • Third, SIMUSAFE will study the behaviours and responses of road users driving, riding, and walking under high-risk situations and impairment conditions.

OBJECTIVES

Model Development and Data Collection

  • An Actor Model of each type (car, pedestrian, two-wheeler) integrating neurometrics and aggregated vehicular/environmental data from naturalistic driving and simulators for identification and representation of driving patterns and computation of risk metrics.
  • Neurometric indexes of risky attitudes and behaviours based on physiological parameters (HR/HRV, EMG, EEG; EOG, ECG and GSR) jointly with contextual information (e.g., Sleep duration/quality, Activity intensity, Weather, Noise). Will comprise risk perception, awareness, attention and decision-making.
  • Integrated Data Collection Module for the filtering of raw data signals and Actor Model descriptor computation with connectivity to cloud-based infrastructure.
  • A quantified risk-taking and risk potential metric for biometric/vehicle data based on the multi-agent model and its equivalent for a simulated virtual driver.
  • Identification of ADCs (risk-perception, awareness) and quantized risk assessment for each class of actor, accordingly with measured data and possible interactions with others actors/environment and its own conditions. Module for observed data incorporation from biometric/vehicle sensors into the simulated agent parameters, such that the behaviour can be reproduced in simulation environment in large-scale.

Accurate Road User Simulation and Integration with Naturalistic Driving Tests

  • Realistic MAS models for driving and traffic simulators able to represent pedestrians, two-wheelers and standard car in traffic context as a dynamic system.
  • Modular API for automotive, two-wheelers or pedestrian simulators, and other actor classes integration.
  • Distributed server-client infrastructure for multi-driver / multi model simulation and DAI based simulation of the various entities of the traffic environment.
  • Modular API for the integration of biometric and vehicle sensors into a simulation module.
  • An in-built automated data analysis module based in Multi-Scale Entropy Analysis (MMSE) to determine relevant descriptors among the data flow produced by the (real/simulated) sensors of the platforms.
  • A methodology for raw data correlation between simulators and naturalistic driving tests based on PCA.
  • A module for data incorporation of measured Actor Model into the Simulated Agent-Based Model such that the behaviour can be reproduced in the simulation environment in large-scale.
  • Analysis and test tool which will reproduce standard test scenarios such as NHTSA pre-crash scenarios and real-world scenarios.

Social Impact

  • Extraction tool of possible relevant factors from environment, drivers (individual or global factors), and other attributes after a test session.
  • Develop interventions (training, regulation) on identified sources of events of interest (near-collisions, traffic jams, infractions) with data analytics tools and experts (psychologists, instructors, traffic authorities)
  • Dissemination events in the form of workshops (14), conferences (54) and fairs (21) involving researchers and stakeholders