An approach for handling uncertainties related to the composition of vehicle fleets in - PowerPoint Presentation

An approach for handling uncertainties related to the composition of vehicle fleets in traffic simulation experiments with automated vehicles Johan Olstam, Fredrik Johansson, Peter Sukennik

Agenda Background & Aim Overview of approaches for simulation of automated vehicles Challenges for traffic simulations of AVs A way to deal with heterogeneity of AV behavior A way to deal with uncertainties w.r.t. evolution of AVs Short on implementation in Vissim Conclusions and next steps

Classification of automated vehicles SAE level Short description 0 – No automation Full-time performance by a human 1 – Driver AssistanceAssistance system of either steering or acceleration/deceleration 2 – Partial AutomationAutomation of some parts of the driving task3 – Conditional Automation Self driving but driver responsible and required to intervene if necessary4 – High AutomationSelf driving in some environment – driver not responsible5 – Full Automation Self driving everywhere

Large expectations, but deployment of automated vehicles will not develop in a perfectly linear transition phase to 100% penetration,will require sharing roads among diversely equipped road users (probably for a long time),will be the result of technological progress, market development, regulation, and urban mobility policy making. 4

Aim How can we investigate AVs impact on traffic performance in a sound and comprehensive way? Traffic simulation should be able to give answers But how? 5

Summary of state-of-the-art of traffic simulations of AVs Most simulations considers separate AV-functions as ACC or CACC Most simulations assumes that all AVs behave the same Some simulations consider different types of AVs but most of them do not consider that the different types of AVs might coexist Most simulation experiments focus on motorways or multi lane roads without interaction with active modes Large variation in traffic performance estimations from previous simulation investigations

Different approaches for simulation of automated vehicles Simulation of automated driving logic by adjustment of behavioural model parameters in the traffic simulation model. Replacing behavioural models in the traffic simulation model with automated vehicle driving logic models Extending the driving behavioural models with “nanoscopic” modelling of automated vehicles, including simulation of sensors, vehicle dynamics and driving logics.

Challenges for simulation of AVs Limited data on first generation AVs No data on future AVs Transition towards full automation will be long Large uncertainties w.r.t. Driving behaviour of AVs Evolution of AV technology and penetration ratesBehaviour of other road users in response to AVs8

How to handle the uncertainties? Scenarios with consistent assumptions Conceptual modelling of AV capabilities wrt perception, anticipation, driving logic rather than detailed modelling of a specific AV-function Sensitivity analysis

Stages of coexistence Introductory: Conventional vehicles still in majority. Automated driving significantly constrained by limitations (real or perceived) in the technology. Established: Automated driving established as an important mode in some areas. Conventional driving still dominates some areas due to limitations (real or perceived) in the technology. Prevalent: Automated driving is the norm, but conventional driving is still present.10

Hierarchical specification of AV driving behaviour The level of automation is specified in two steps: AV class Basic Intermediate AdvancedDriving logic (for different road environments) Rail-safeCautiousNormalAll-knowing11

AV classes Basic: SD only in one directional traffic with physical separation to active modes. No dedicated devices for vehicle communication and cooperating functions. Intermediate: SD in structured traffic.May have dedicated devices for vehicle communication and cooperating functions, but are not depended on them.Advanced: SD in most environmentsWill have dedicated devices for vehicle communication and cooperating functions, but are not depended on them.12

Driving logics Rail-safe: Based on the switch principle. Follows pre-defined path. Cautious: Require large gaps; slows down every time its sensors can have blind angles. Normal: Similar to a human driver but with the augmented (and/or diminished) perception due to sensors.All-knowing: Perfect perception and prediction of the behaviour of other road users. Capable of offensive driving whenever needed, without causing accidents.13

Relation AV-class and Driving logics for different Operational Design Domains 14 Road type Driving logic: Rail-Safe (RS), Cautious (C), Normal (N), All-Knowing (AK), Manual (M) Basic Intermediate Advanced Motorway C N AK Arterial C C / N AK Urban street M C N Shared space M RS / M C

Example of penetration rates & shares 15 AV penetration Basic AV shareIntermediate AV share Advanced AV share Introductory 10-40 70-100 0-30   Established 30-70 0-20 80-100 0-10 Prevalent 60-90   20-80 20-80

Implementation in traffic simulation Implementation was done for Vissim Based on the description of each driving logic we assessed whether the behavioral model parameters are likely to be unaffected, decrease or increase The Brick wall stop distance principle was implemented for the Rail safe and Cautious driving logics Varying perception of number of surrounding vehicles“Calibration” of some of the behavioral parameters were conducted based on field tests with three ACC/CACC vehicles.16

Example from Implementation in Vissim   driving logic   model parameter** rail safe cautious normal all knowing following behavior Wiedemann 99 CC0 - Average desired standstill distance def def def smaller CC1 - Desired time headway def/higher* def/higher* def smaller CC2 - Following variation (oscillation) def/smaller def/smaller smaller smaller CC3 - "Delay" for triggering deceleration def/higher def/higher def def CC4 - Negative 'following' threshold Smaller def/smaller def/smallersmallerCC5 - positive 'following' thresholdSmallerdef/smallerdef/smallersmallerCC6 - Influence of distance on speed oscillationdef/smallerdef/smallerdefsmallerCC7 - Acceleration level at oscillationdef/smallerdef/smallerdef/smallersmallerCC8 - Desried acceleration at standstillSmallersmallerdefdefCC9 - Desired acceleration at 80 km/hSmallersmallerdefdef* if EABK is on, brick wall stop distance is guaranteed ** see PTV Vissim manual for detailed description   rail safe cautious** normal all knowing parameter for necessary lane change* own trailing vehicle own trailing vehicle own trailing vehicle own trailing vehicle maximum deceleration n.a. n.a. smaller/def smaller/def def smaller/def def higher/def - 1 m/s per distance n.a. n.a. smaller/def smaller / def def def def smaller/def accepted deceleration n.a. n.a . smaller/def smaller/def def def def higher/def *necessary lane change means a lane change which is necessary in order to follow a defined route (it is not overtaking because of higher own desired speed) ** EABD (enforce absolute breaking distance) must be on n.a. = not applicable

Implementation in Vissim (cont’d) Explicit stochastics: Variance in driver behaviour parameter distribution, e.g. desired/maximum acceleration/deceleration Implicit stochastics: Vissim-”internal” stochastic variation that is meant to model the imperfection of human driverssafety distancesdesired acceleration/decelerationestimation uncertainty for braking decisions18

Conclusions and next steps Conceptual descriptions of different types of AV have been developed and implemented in a traffic simulation model Consistent scenarios with respect to technological development, deployment The developed approach will be applied for 5 different use cases around Europe 19

Get in touch! www.h2020-CoEXist.eu 20 @H2020_CoEXist #H2020_CoEXist Johan Olstam johan.olstam@vti.se OR CoEXist Coordination: Wolfgang Backhaus w.backhaus@rupprecht-consult.eu

Johan Olstam johan.olstam@vti.se The sole responsibility for the content of this document lies with the authors. It does not necessarily reflect the opinion of the European Union. Neither the EASME nor the European Commission are responsible for any use that may be made of the information contained therein . #H2020CoEXist @H2020_CoEXist