MULTI-AGENT SYSTEM AND DATA ANALYTICS MULTI-AGENT SYSTEM AND DATA - PowerPoint PPT Presentation

SAFETY AND MOBILITY APPLICATION WITH SAFETY AND MOBILITY APPLICATION WITH MULTI-AGENT SYSTEM AND DATA ANALYTICS MULTI-AGENT SYSTEM AND DATA ANALYTICS Monty Abbas, Virginia Tech VT-SCORES (Qichao Wang and Awad Abdelhalim) Data Discovery for November 16-17, 2017 1

Background: Multi-agent System Background: Multi-agent System Goals System Agents 3

Vision Vision Calibration s Safety Simulation API Simulation Kit s NDS Varying Traffic Vehicle s Network Data Behavior Maneuvers Safety Heat Map from Simulation (speed, acceleration, steering, etc.) Agent s learning process Straight travel, constant speed Straight travel, Straight travel, accelerate decelerate Other states Steer, accelerate Steer, decelerate Steer, constant speed Safety Heat Map from SmarterRoads Agents Library: Type and Frequency 4

Naturalistic driving behavior: event data Naturalistic driving behavior: event data • Training input: traffic states and actions • Training output: acceleration and steering • Input variables discretized using fuzzy sets • Continuous actions are generated from discrete actions • Produces Individual agents combining acceleration and steering behavior 5

Learning Techniques Learning Techniques Policy P State S Diagram State 1 State State 2 State 3 Other states State 6 State 4 Action State 5 • Using actor-critic Reinforcement Learning 6

Agent 1: Eta* Agent 1: Eta* 7

Agent 2: Virginia* Agent 2: Virginia* 8

Utilizing SmarterRoads Crash data Utilizing SmarterRoads Crash data 9

The Matrix has you The Matrix has you 10

Summary Summary • Possibilities and Impact • Develop a library of agents (including disadvantages population) • Calibrate agent distribution in simulation to replicate safety performance at a site • Evaluate potential improvement strategies for public safety and mobility 11

Summary Summary • Innovation • Combined car-following and lane changing for normal and safety-critical driving • Better, more accurate modeling that can accommodate disruptive technology (e.g., CAV) • Can replicate existing safety performance (crashes) and mobility (congestion) in a region 12

“…I don’t know the future. I didn’t come here to tell you how this is going to end; I came here to tell you how it is going to begin!” Thank you! Thank you! 13