Next Generation of Adaptive Traffic Signal Control Pitu Mirchandani ATLAS Research Laboratory Arizona State University NSF Workshop Rutgers, New Brunswick, NJ June 7, 2010 Acknowledgements: FHWA, ADOT, NSF
Outline 3-part talk Current Responsive Traffic Control Practices & Issues Real-time Adaptive Control RHODES - Current RHODES - Next Generation RHODES - Future with IntelliDrive Conclusions Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
CURRENT PRACTICE TRAFFIC RESPONSIVE SYSTEMS UTCS (Urban Traffic Control System, FHWA, US, 1070 s) 2 nd and 3 rd generation systems have adaptive features. SCOOT (Split, Cycle, and Offset Optimization Technique, UK, 1970 s) Monitor traffic volumes and frequently (every few cycles) develop a new plan based on TRANSYT New detectors needed downstream to measure traffic profiles SCATS (Sydney Coordinated Adaptive Traffic System, Australia, 1970 S): A degree of saturation is measured at each approach Cycle time is increased when average saturation increases, and Splits are allocated in proportion to saturation Adjacent intersections are grouped when cycle times are nearly same, or ungrouped for different cycle times demand. Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
CURRENT PRACTICE TRAFFIC RESPONSIVE SYSTEMS OPAC (Optimization Policies for Adaptive Control, US, early 1980 s by Gartner et al.) First to move away from traditional plans Upstream detectors measure approach load For a given time horizon, various combinations of green phases are analyzed, and optimum durations are selected based on implicit enumeration. Current RHODES optimization model uses these ideas. Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
CURRENT PRACTICE ISSUES Few jurisdictions use adaptive control mainly because They are hard to implement Require additional sensors Improve performance only when system is under saturated Next generation adaptive control must respond to above concerns. But note that there is always a capacity for a signalized network, and when the load is increased above this capacity there will be unbounded queues no matter what one does. What the next generation control will do is increase this capacity as much as possible. Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Framework For Real-time Decision Systems Equipment Processing Data Gathering data flow Decision System Sensors Sensor media Feedback & decisions Real-world We will keep coming back to this! Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
What is an Adaptive Control System? It is necessarily a Feedback Control System that Adapts data Real-time Control System Measurements: monitoring state of system Feedback & decisions Actual System Controls: Actuators Signals. Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
General Adaptive Control Architecture Decision/Control Algorithms (using desired objectives) We will keep referring to this architecture Exogenous inputs Model Optimization Decisions/Controls u(t) Real-World Systems x(t) Model Estimation Estimator/Predictor Current Adaptive Control y(t) Comm. delay Measurements (Latency delay) x(t) Sensors OUTPUTS (states of the system.) Measurement noise RHODES RHODES/NG RHODES/VII Conclusion
Problems and Issues with Current Traffic Management Paradigms Little recognition that traffic state is a non-stationary stochastic process E.g.: A traffic plan (cycles, splits and offsets) assumes that the process is stationary Traffic Adaptive requires constant monitoring of traffic this is the cost of adaptive performance E.g., RHODES (our adaptive traffic control) does not use plans but assumes that some real-time information is available all the time Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Traffic- Adaptive Signal Control System Adaptive Traffic Signal Control System raw data Measurements: detectors &signals Actuators: signals Feedback & decisions Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Quality Attributes of an Adaptive Traffic Signal Control System? Responsiveness: How fast does it respond to changes in traffic conditions? (including incidents and special events) Feedback Philosophy: Is it reactive? (the vanilla version) Is it proactive? (the sundae version) Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Open-loop, Reactive & Proactive (Illustration in following a trajectory) Position Actual Trajectory Reactive Open Loop Proactive Time Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Adaptive (Real-Time Proactive) Traffic Control Explicitly recognizes that traffic state is a non-stationary stochastic process Requires prediction of short-term future based on current conditions and controls Especially useful for non-recurrent traffic conditions and major incidents Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Scenario Hierarchical Architecture for RHODES ADAPTIVE TRAFFIC MANAGEMENT Origins/Destinations Historical/Infrastructure Data Destinations/Origins Network Load Control Network Loads Target Timings Network Flow Control Intersection Control Actual Timings Current Capacities, Travel Times, Network Disruptions (minutes/hours/days) Regional Network Feedback Platoon Flow Prediction (minutes) Network level Feedback Vehicle Flow Prediction (seconds) Network Load Estimator/Predictor Network Flow Estimator/Predictor Intersection Flow Estimator/Predictor Control Signal Traffic Signal Activation Intersection Feedback Control ATIS Reference: Head, Mirchandani, Sheppard, 1992 Actual Travel Behavior and Traffic Detectors and Surveillance y(t) Measurements Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Simplified Architecture for RHODES Data collection and prediction of queues and arrivals raw data processed data Control Selection Feedback & decisions Detectors, traffic signals, and communication Control Actions (phase durations) Counts Stop-bar Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
RHODES: THE GENERAL PROACTIVE IDEA Decision/Control Real-time Control/Decision Algorithms (using desired objectives) PREDICT Model Optimization Decisions/Controls u(t) Exogenous inputs Real-World Traffic Systems x(t) CAPRI PREDICT Model Estimation Real-time Estimator/Predictor Current Adaptive Control RHODES y(t) Comm. delay Measurements (Latency delay) x(t) Sensors OUTPUTS (traffic volumes, speeds, queues, air quality, etc.) Measurement noise RHODES/NG RHODES/VII Conclusion
PREDICTION & CONTROL IN RHODES PREDICT arrivals & queues CONTROL ALGORITHMS (CAPRI) TURN RATIOS TRAVEL TIMES DISCHARGE RATES detectors state of traffic network Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
RHODES: INTERSECTION PREDICTION UNDER RHODES CONTROL ca rdetector Me Traffic Mgt RHODES Evacuation Dynamic Flows Cases Future
RHODES: INTERSECTION PREDICTION And... PREDICTIONS! Next second A little later R.3 1.3.6.3.3.3.3.3.3 T 1.5 2 1.5 2 1.5.5.5.5.5.5 L.2 1.2.4.2.2.2.2.2.2 1 2 3 4 45 46 47 48 49 50 51 52 Time Me Traffic Mgt RHODES Evacuation Dynamic Flows Cases Future
RHODES INTERSECTION CONTROL R T L R T L R T L R T L 1 2 1.6 2.4.3 1.5.2.3.5.2.3.5.2 1 2 3 4 45 46 47 48 49 50 51 52 PHASE ORDER: B-C-D-A-B-C-D-A... B C D A B C.3.5.2.3.5.2.3.5.2 From North From South From West From East Time We can easily compute total delay and stops from this diagram RHODES idea is to change Phase durations to minimize cost. 1 2 3 4 45 46 47 48 49 50 51 52 Time
CAPRI*:INTERSECTION CONTROL LOGIC Effectively, a real-time algorithm that determines: for a given Phase Order A,B,C,D,A,B,C,D... what time durations should be given to Phase A, Phase B,..., etc. allows various objectives (stops, delays, queues) for different classes (cars, buses,...) considers categories of predicted arrivals and their objectives considers a given rolling decision time horizon T, with time increments of D seconds (roll period) * Categorized Arrivals-based Phase Re-optimization at Intersections. Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Performance - Simulation (Atlanta) SAC RHODES Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
RHODES Installations RHODES Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Additional Features - Transit Priority NETWORK FLOW CONTROL SUBSYSTEM APRES-NET arrivals & queues REALBAND INTERSECTION CONTROL SUBSYSTEM PREDICT TRAVEL TIMES TURN RATIOS arrivals & queues CAPRI DISCHARGE RATES detectors Transit/bus Priority (position and weight ) Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Additional Features - Emergency Preemption NETWORK FLOW CONTROL SUBSYSTEM APRES-NET arrivals & queues REALBAND INTERSECTION CONTROL SUBSYTEM DISCHARGE RATES PREDICT TRAVEL TIMES detectors TURN RATIOS arrivals & queues CAPRI Emergency vehicles (phase constraints) Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Additional Features - Emergency Preemption Location of incident reported Shortest route computed based on real-time traffic conditions and given to dispatcher Traffic signals pre-empted based on shortest route from depot to incident Depot Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Additional Features - Rail Preemption Train movement (position and schedule) NETWORK FLOW CONTROL SUBSYSTEM APRES-NET arrivals & queues REALBAND INTERSECTION CONTROL SUBSYTEM DISCHARGE RATES PREDICT TRAVEL TIMES TURN RATIOS arrivals & queues CONTROL ALGORITHMS Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
2 nd Part Current Traffic Control Practices Real-time Adaptive Control RHODES - Current RHODES - Next Generation RHODES Next Generation with IntelliDrive Conclusions Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
The performance of RHODES is directly related to the accuracy of its queue estimates Parameters which affect this accuracy: Turn Proportions Proportion of vehicles on an approach which turn left, turn right or proceed through the intersection Queue Discharge Rates Rate at which vehicles leave an intersection, dependent upon the number of available lanes and the movement involved Link Travel Times RHODES Input Parameters Time taken by a vehicle to traverse the distance from an upstream peer intersection to a point downstream Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
RHODES Self-Adaptive Traffic Signal Control Self-adaptive Traffic Signal Control Next Generation Control Systems Incorporate algorithms that automatically update critical RHODES parameters based on available data Benefits Performance of RHODES will be further improved Significant reduction in calibration and fine-tuning Eliminates the need to update parameters periodically Data and computed parameters will be available to agencies for other purposes, such as regional planning Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
PREDICTION & CONTROL IN RHODES-NG PREDICT arrivals & queues CONTROL ALGORITHMS (CAPRI) TURN RATIOS TRAVEL TIMES DISCHARGE RATES detectors state of traffic network Issues Framework RHODES Evacuation Dynamic Flows Cases Future
INTERSECTION CONTROL SUBSYTEM PREDICT arrivals & queues CAPRI TRAVEL TIMES TURN RATIOS FINITE HORIZON DYNAMIC PROGRAM DISCHARGE RATES detectors Issues Framework RHODES Evacuation Dynamic Flows Cases Future
INTERSECTION CONTROL SUBSYTEM PREDICT arrivals & queues CAPRI TURN RATIOS DISCHARGE RATES TRAVEL TIMES GENERALIZED LEAST-SQUARE ESTIMATION detectors Issues Framework RHODES Evacuation Dynamic Flows Cases Future
INTERSECTION CONTROL SUBSYTEM PREDICT arrivals & queues CAPRI TURN RATIOS TRAVEL TIMES DISCHARGE RATES detectors Issues Framework REAL-TIME PLATOON TRACKING RHODES Evacuation Dynamic Flows Cases Future
INTERSECTION CONTROL SUBSYTEM PREDICT arrivals & queues CAPRI TRAVEL TIMES TURN RATIOS THIS IS SUPPORTED BY AN ON-GOING FHWA CONTRACT DISCHARGE RATES detectors Issues Framework MONITORING ESTIMATED QUEUES & DETECTOR OCCUPANCIES RHODES Evacuation Dynamic Flows Cases Future
Adaptive Turn Proportions Auto configuration based upon intersection geometrics/phasing Auto adjusts to reflect actual turn proportion variability Simulation results show an improvement in performance
Adaptive Turn Proportions Auto configuration based upon intersection geometrics/phasing Auto adjusts to reflect actual turn proportion variability Simulation results show an improvement in performance
Adaptive Turn Proportion Sample Results Approach-1-through movement turning proportion 1.2 1 0.8 0.6 0.4 0.2 0 algorithm's prediction three cycle's average 241 754 1187 1664 2155 2600 3269 3787 4433 5030 5929 time Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
RHODES Self-Adaptive Traffic Signal Control RHODES Self-adaptive Traffic Signal Control responds to these issues 1. Changing short-term demand and in the long run will automatically equilibrate with network flow changes (bi-level dynamic network equilibrium) 2. Saturated traffic conditions (up to a maximum capacity) 3. Accepts and integrates data from IntelliDrive systems Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
RHODES Self-Adaptive Traffic Signal Control Self-adaptive Traffic Signal Control Automatically recognizes various operating regimes Queue size Residual queues keep exploding (over saturation) Usually no residual queues Residual queues described by steady-state distribution Traffic load Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
RHODES Self-Adaptive Traffic Signal Control Self-adaptive Traffic Signal Control Automatically recognizes various operating regimes Queue size Load info provided from upstream to downstream (usually no residual queues) (residual queues described by steady-state distribution) Illustrated this earlier Traffic load
RHODES Self-Adaptive Traffic Signal Control Self-adaptive Traffic Signal Control Automatically recognizes various operating regimes Queue size Load info provided from upstream to downstream (usually no residual queues) Illustrated this earlier Residual queues keep exploding (over saturation) (residual queues described by steady-state distribution) Queue build info provided from downstream to upstream* Traffic load [* info on end of queue to prevent spill-back at upstream intersection]
RHODES Self-Adaptive Traffic Signal Control Additional benefit: performance monitoring Queues, delays and travel times, Level of congestion operational regimes Unsaturated Saturated but stable Over saturated (unstable) Route travel times
RHODES Next Generation w/intellidrive Need DATA FUSION to predict demand for various signal services RSU RSU OBU OBU RHODES with/intellidrive Integration Scheduling of multiple preemption/priority requests Data exchange occurs between On Board Units (OBU), Road Side Units (RSU), the signal controller and RHODES. (Currently using DSRC) OBU RSU NG-RHODES will provide appropriate service for various classes of vehicles Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
3 rd Part Current Traffic Control Practices Real-time Adaptive Control RHODES - Current RHODES - Next Generation RHODES Next Generation with IntelliDrive Conclusions Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Concluding Remarks on Next Generation Adaptive Traffic Control Improvement in traffic performance: responds to recurrent congestion responds to near oversaturation responds to non-recurrent conditions and incidents (through monitor, learn, predict and optimally respond strategy) Decrease in traffic operations/planning effort operators need not time signals periodically planners and traffic engineers can concentrate on smaller number of scenarios Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Concluding Remarks NEAR FUTURE: Special vehicles will be identified via transponders and detectors, e.g.: Emergency, Transit, HAZMAT, using IntelliDrive structure Traffic signals will provide appropriate signal service by scheduling the service within the given time horizon FAR FUTURE: Every vehicle will be tracked. Every vehicle will be require and be provided appropriate service and treated with appropriate priority. Signals will provide in-vehicle signal and controls ( STOP or you will have an accident ). Safety will improve. Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion
Thanks for your attention Questions???
Simulation Results Without Turning Proportions Estimation Vehicle Vehicle Vehicle minutes Travel time Avg. speed Avg. stop Miles Trips Delay (Sec/Veh-Trip) (MPH) (Per Trip) time Period 1 3059 6711 3576 80.3 20.4.7 Period 2 5474 12304 5737 75.1 21.3.7 Period 3 8002 18369 8292 73.0 21.5.6 With Turning Proportions Estimation Vehicle Vehicle Vehicle minutes Travel time Avg. speed Avg. stop Miles Trips Delay time (Sec/Veh-Trip) (MPH) (Per Trip) Period 1 3058 6712 3039 75.4 21.7.6 Period 2 5467 12284 4760 70.4 22.8.6 Period 3 7979 18315 7386 70.1 22.4.6 Current Adaptive Control RHODES RHODES/NG RHODES/VII Conclusion