Next Generation Simulation (NGSIM)
-Improved Simulation of Stop Bar Driver Behavior at Signalized Intersections-
Los Angeles, CA and Atlanta, GA Data Sets

Working Papers

Summary

Introduction

Improving the ‘realism’ of traffic simulation models has been in progress ever since the first models were developed in the 1950s. TRAF, an integrated traffic simulation model was developed by the FHWA; it did not involve any new model development but was based on improving the best existing traffic simulation models. INTRAS was the best microscopic freeway simulation model, but due to some of the inherent assumptions in its car following model (for e.g., not considering start-up delay of stopped vehicles, not considering dual behavior of traffic in congested and non-congested conditions, etc.), it was unfit for modeling stop-and-go conditions on freeways. CARSIM was developed to address those deficiencies and offer other realistic features. It was then validated at both microscopic and macroscopic levels using trajectory-data collected by Treiterer (co-author of this paper) at Ohio State University.

Features and Description of CARSIM

CARSIM consists of two modules: a car-following algorithm and an acceleration algorithm. For the former, the following features were included in its development:

·          Vehicles’ acceleration and deceleration rates were kept within reasonable ranges similar to what can be observed in actual conditions

·          Delay in response (reaction time) of the driver of a following vehicle to the lead vehicle was taken into consideration

·          Startup delay of a stopped driver was taken into consideration

·          Dual behavior of traffic in congested and non-congested conditions was also taken into consideration

·          Varying reaction times for an individual driver and different reaction times for different drivers were taken into consideration.

Vehicles in the simulation were generated one at a time with each vehicle having a different set of attributes. The traffic mix is a user-specified input variable. The position of each vehicle in the system was scanned at every one second and its position and velocity were computed using basic kinetic equations. Depending on whether a vehicle was traveling below, at or above the desired speed (or speed limit), accelerations of 0 ft/s² to a ‘comfortable deceleration rate’ were applied to update vehicle position and speed.

The acceleration algorithm was such that it computed acceleration/deceleration rates for vehicles in 1-second intervals while satisfying all safety and operational constraints. Five values of acceleration (A1 through A5) were calculated based on certain situations:

·          A1: The following vehicle is moving but has not reached its desired speed (or speed limit): It is obtained from a table (Table 1 in the paper) for a given type of vehicle and speed.

·          A2: The following vehicle has reached its desired speed (or speed limit): CARSIM assumes that a driver will try to reach his desired speed (or the speed limit) as fast as possible while satisfying all safety and operational constraints.

·          A3: The following vehicle was stopped and has to start from a standing still position: A3 will depend on start-up delay (1-3 seconds), which in turn depends on driver reaction times. CARSIM assumes less than 20% of drivers have a reaction time of 0.68s (rounded to 1s in the model); most drivers will wait for 2 seconds before moving. The acceleration rate for vehicles starting from a stopped position is 2 ft/s² for passenger cars and 1 ft/s² for trucks for the first second of motion. Thereafter, the car-following algorithm is used to compute position, speed and acceleration.

·          A4: The following vehicle’s performance is governed by the car-following algorithm while space headway constraint is satisfied: A basic equation based on buffer space between two vehicles and the length of the leading vehicle was used to compute A4. This buffer was 10ft for less dense conditions and 5-7ft for more dense conditions.

·          A5: The following vehicle is advancing according to the car following algorithm with the non-collision constraint: A5 is based on brake-reaction time of the (following) driver, velocity of the leading car and maximum deceleration rates of the leading and following vehicles.

Once A1 through A5 are calculated, the ‘proper’ acceleration rates (min. of A1 through A5) and ‘proper’ deceleration rates (which could be A2 or A5 or ‘comfortable deceleration rate’ depending on some conditions) are chosen.

The ‘Brake Reaction Time’ (varying from 0.4 to 1.5s in 0.1s increments) is a measure of how drivers would react when they encountered kinematic disturbances. Each driver was assigned one value of BRT when he entered the system. Vehicles in dense conditions (>60vpm) were assigned lower BRTs and vice versa. Younger drivers were assigned lower BRTs while older drivers were assigned slightly larger values. For truck drivers, minimum BRT was assumed to be 1s.

Validation

Four platoons of vehicles with differing characteristics were selected from OSU’s trajectory data (mentioned in the introduction) for validation. The data consisted of photographs taken at every second; locations of vehicles were estimated to be accurate within 0.5ft and speeds were determined with an error of 1mph. Spacing, headway and position data was provided for all vehicles.

Validation at microscopic level was performed by observing average speed and location of each vehicle in CARSIM and comparing them to values obtained from field data for that vehicle. For macroscopic-level validation, average speed, volume and density from field data was compared to CARSIM’s output.