Roadrunner+

An Autonomous Intersection Management Cooperating with Connected Autonomous Vehicles and Pedestrians with Spillback Considered

Requirements:

ortools (pip install ortools)
matplotlib (pip install matplotlib) numpy (pip install numpy)

Platform:

Run with SUMO 18.0, python 3

Quick walkthrough with files:

1. main.py

• Start SUMO with gui or without gui (Line 146sumo-gui/ sumo in checkBinary)
• Generate intersection information (Line 149, call gen_data.sh to generate intersection infos)
• Generate traffic with given arrival rate and random seed (Line 153, call the function in gen_route.py)
• "Usage: python main.py <arrival_rate (0~1.0)> <is_slowdown_control T/F>"
• The is integers from 0~3, used in (Line 384 in IntersectionManager.py)
  • 0: algorithm in Roadrunner+ (it was temporarily named IcaccPlus)
  • 1: ICACC (which didn't implement pedestrian)
  • 2: FCFS reservation (also consider pedestrian)
  • 3: First-come first-pass (FCFS_no_reservation, which didn't implement pedestrian)

2. config.py

• Records all the configurations
  • LANE_NUM_PER_DIRECTION: number of lane per direction in a four-legged intersection
  • MAX_SPEED/TURN_SPEED: the speed defined for Roadrunner+ to control near the intersection (in m/s)
  • SUMO_TIME_ERR: this is to calibrate the time difference between SUMO simulation and Roadrunner+ algorithm (Look into this when collision happens at speed control near the intersection)
  • *_LEN: length of each zone
  • LANE_WIDTH: the width of lane (this number needs to match the setting of SUMO network)
  • HEADWAY: a defult headway between cars (3 meters)
  • CCZ_*_LEN, MAX_ACC: the detailed setting for the cuise control zone (i.e., the length needs to be smaller than CCZ_LEN)
  • CAR_*: the length of the cars
  • N_TIME_STEP: time of simulation that SUMO has to conduct (in seconds in SUMO time)
  • TIME_STEP: size of time step that SUMO takes for the simulation
  • PEDESTRIAN_POSSIBILITY: probability of generating a pedestrian in each time step (the pedestrian is not printed in SUMO)
  • PEDESTRIAN_TIME_GAP: the time gap that researved for the pedestrian

3. data/

• Record the settings for the SUMO simulation, including the network (.net.xml)
• Generated traffic information (.rou.xml) is saved here

4. gen_route.py

• The traffic generating function (generate_routefile) is called in main.py
• The traffic data is saved in data/

5. inter_info/

• The recorded safe gap data within the intersection
• Files: lane_info*.json. The number is the lane number per direction of the intersection
• Files are generated with the programs in gen_intersection.
• Needs to generate the according files before the simulation starts (i.e., calling main.py).

6. inter_length_info/

• The recorded trajectory length within the intersection for each turn and lane. (The information is used for lane advising)
• Files: lane_info*.json. The number is the lane number per direction of the intersection
• Files are generated with the programs in gen_intersection.
• Needs to generate the according files before the simulation starts (i.e., calling main.py).

7. advise_info/

• The recorded blocks that needs to be reserved in the intersection. (The information is used for lane advising)
• Files: advise_info*.json. The number is the lane number per direction of the intersection
• Files are generated with the programs in gen_intersection.
• Needs to generate the according files before the simulation starts (i.e., calling main.py).

8. gen_intersection/

• Programs that generate files in 5-7.
• Use $make to compile the .cpp files (These .cpp generates the basic information of the intersection, regardless the config.py settings)
• Call "$bash gen_data.sh" to convert the raw information to a usible data with config.py settings.
• After running make and bash, the results should be found in 5-7.

9. milp.py

• The MILP formulations for each scheduling algorithm
• Note that FCFS is implemented with Roadrunner+ algorithm (IcaccPlus) by sending one car at a time
• Note that FCFS_no_reservation is implemented with ICACC by sending one car at a time (because it force the later group to wait for the first-arrived group)
• Note that FixedSignal is deprecated and replaced with a more reasonable traffic-signal model

10. Cars.py

• The car control stages and behaviors, mainly the cruise control behaviors

11. IntersectionManager.py

• The "main" of Roadrunner+, which includes the lane advising, scheduling, communicating with cars, and pedestrian handling