Traffic Flow Simulation Car-Following Model By: Ittinop(Pun) Dumnernchanvanit Introduction: What is done in this project? ◦ Simulate traffic following each individual car. ◦ Use AI to simulate driver’s behavior on road ◦ Observe and analyze traffic phenomena Why car-following model? ◦ Traffic is extremely complex and most phenomena are non-linear and cannot be solved easily and accurately through equations. Example uses for simulation: ◦ Can help maximize traffic flow. For instance, determine the most efficient automated red light, green light pattern. ◦ How far to put a warning sign for road block. ◦ Help in choosing between stop sign/red light green light at specific intersection Phenomena Simulated: Shockwave Road block Cutting in front Platoon System of four way and three way intersections Mechanisms behind the simulation Speed Density Approximation Mechanisms behind the simulation Acceleration: Mechanisms behind the simulation Angular movement calculated using turn radius. ◦ This way we can turn without worrying about speed Mechanisms behind the simulation: Angular movement Mechanism behind the simulation: Angular movement Angular movement: Mechanisms behind the simulation: Turn radius Turn radius calculation: (for future improvement) ◦ 1. Find intersection using y = mx+b etc. ◦ 2. Find distance from lane end to intersection ◦ 3. Find angle 3 ◦ 4. Find turn radius Mechanisms behind the simulation how to show vehicle with its direction take car to center, then use rotational matrix, then take it back Project code composition: Car Lane Creator main Car Variables: ◦ ◦ ◦ ◦ ◦ ◦ ◦ (x,y) Direction Max speed Max acceleration Max brake Lane waypoints Car Methods: ◦ react() –determines acceleration and angular movement. Basically that is how real world driver control car, pedal/brake for acceleration and wheel turning for angular movement. ◦ move() – move the car according to acceleration and angular movement. ◦ getFrontCar(), getBackCar(), getBackMostCar() Lane Variables ◦ ◦ ◦ ◦ ◦ Position Width Direction Leftlane, rightlane Start, end Methods ◦ getDirection() ◦ insertAdjacentLane(Lane* leftlane_raw, Lane* rightlane_raw) ◦ isEqual(Lane* a) Creator Works like car factory that spit out car on to lane from some specific point. Spit out if no car with in a specific distance starting_speed: ◦ = max_speed*(1-min_d/d); Adjust to different density automatically and will not over produce. Can adjust density using this. Creator Variables: ◦ ◦ ◦ ◦ Waypoints, endland transitions Distance between cars Starting speed Chance to produce etc. Methods ◦ closeByCar() test if there is car near by the creator object (Can adapt do different density) ◦ createCar() Main Set-up the system ◦ Build lanes, and creators/or cars loop through time steps ◦ Run the car ◦ Record the results Assumptions • All units in meters and seconds • chose 0.1 sec for time step because human reaction is 0.2 sec Max speed: 65 kmph Max acceleration: 3.79 m/s ◦ (~7.1s 0-60mph) Minimum distance between car: ◦ 7m from center to center, or around 2-3m between car. Max Flux Derivation What should max flux be? Steady State Movie Max Flux Data Distance between cars (m) 24 19 14 9 4 flux (n/s) 1.06 1.18 1.27 (calc. = 1.285) 1 0.66 Shockwave Traveling disturbance in distribution of cars on road. Usually backward motion Shockwave Video: Platoon This is an idea to group vehicle in to platoons to increase the capacity of road. ◦ This allow cars to be closer to each other. This will need smart car that can be driven by artificial intelligence Platoon Video: Platoon Data comparison: (Assume that on one of the lane, there is 50%/50% chance that creator will produce platoon or car.) Platoon No Platoon average final time (s) 24.1601 27.174 flux (n/s) 1.73 1.27 Road Block: How it is done: ◦ Car object contain pointer to object targetlane and lane ◦ Why targetlane? vehicles will also check other vehicle’s targetlane in their loop so they can recognize incoming car from another lane and yield for it. Road Block: Algorithm ◦ At road block, vehicle slow down and tries to cut in front of another vehicle. ◦ distance to back car and to front car in target lane vehicle need to cut is set ◦ when vehicle is set to change lane, it turn and run toward the lane and then turn the wheel back when it is in the middle and Something to keep in mind when looking at data Time is counted from entering system to exiting system. If flux is low, it might means traffic jam might propagate much longer than system which means the car would have waited much longer outside the system than the case with less flux Road Block Movie No sign, see block at around 100 Road Block Movie Sign at 100, see actual block around 200 Road Block Data: No Warning Sign Production average flux (n/s) distance final time (s) 24 70.7099 0.37 34 41.8327 0.366667 44 36.5018 0.37 Warning Sign average flux (n/s) final time (s) 51.7759 0.36 28.6344 0.426667 23.2858 0.423333 Cut in Front: How it is done Driver looks to another lane to decide whether it is worth to cut in front. Then look to the back to determine if it is possible to do so Cut in Front: How it is done Judgment criteria driver use for front car: ◦ coefficient*( (speed of front car in our lane)*time +distance to front car in our lane )) ◦ (speed of front car target lane)*time +distance to front car in target lane ◦ where time is any set time, depending on driver’s experience. ◦ coefficient allow us to set how much we want the driver to cut Cut in Front: How it is done Then look at back car ◦ driver look at speed of back car and distance to back car. ◦ driver knows the amount of time he will use to cut in front. ◦ simple algorithm used is just, (coefficient*distance) >( (back car speed)*(cut time)) Cut in Front: Movie Show outside Cut in Front: Data (5 lanes), del_t = 1.0s Distance = 14 m Coefficient Distance = 54 m flux (n/s) Number of cuts per car 1.1 average final time (s) 48.9568 flux (n/s) Number of cuts per car 1.69195 average final time (s) 38.8109 1.45 0.95 0.22807 1.5 48.3438 1.49 1.05593 38.7677 0.95 0.157895 2.0 46.417 1.48667 0.44843 38.7014 0.95 0.101754 3.0 45.8578 1.49333 0.176339 38.7028 0.95 0.0982456 5.0 45.7897 1.49 0.152125 38.7021 0.95 0.0982456 Keep in mind: Note that there are so much more variables such as car density we can manipulate and these behaviors might change totally. Complex Road System: System of four lane with three lane attached to it on the east Cases: ◦ Red/green light ◦ all-way stop sign Complex Road System: How it is done Waypoints: ◦ the way point build and given to car creator ◦ makes sense because driver usually knows where he is going to go from the start. (most of the time) ◦ In this project, the waypoints are different lanes the car will go through before exiting Complex Road System: How it is done Red light/Green light set up ◦ Red light are built into lane class. Basically, car on the lane check if it is turned on, if so stop at the light if front car is farther than the lane end. ◦ Set red light in main class. Set repeating pattern using fmod() ◦ Assume no left turn Complex Road System: Red light/Green light Video Show outside Complex Road System: How it is done All way Stop signs in this project was extended from red light code. Check area in the middle and open green light to let some car in temporarily. Complex Road System: All-way Stop Signs Video ◦ Show outside Complex Road System: Data Red light/green light Production average flux (n/s) Distance final time (s) 7 102.126 1.87143 15 97.4733 1.87143 30 90.122 1.81429 50 83.9205 1.67143 Stop signs average final time (s) 110.135 104.56 92.4511 74.2911 flux (n/s) 0.485714 0.614286 0.642857 0.642857 Complex Road System: Analysis So traffic lights are better than stop signs in traffic flow at high flux. They have around the same efficiency at low flux. This is the reason why why we see all-way stop signs in areas with less traffic. The End: Thank you for Listening!