MOTORCYCLE’S BEHAVIOR MODEL AND SIMULATION BASED ON MARKOV DECISION PROCESS IN MIXED TRAFFIC

• Main Author: Mardiati, Rina
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42746
• Institution: Institut Teknologi Bandung

Modelingtraf?cisanalternativewayofanalyzingtherealphenomenaoftraf?c. In order to gain a better understanding of traf?c systems, a microscopic scale could provideabettersimulationresultandprovideagreatdealofusefulinformationfor policymakerstodeterminethebestsolutionfortransportationproblems. Thereare several methods that can be used for developing motorcycle behavior models on a microscopic scale, one of which is Markov Decision Process (MDP). WithMDP,vehiclebehaviorcanbemodeledasasequenceofevents(states)atevery time step. There are four main components in MDP: states, actions, a probability function, and a reward function. The transition of states is caused by execution of actions based on the probability function. In this study, MDP was chosen as the model to describe motorcycle behavior in order to get better simulation results in describing motorcycle behavior, especially in mixed traf?c conditions. In this dissertation, a weighted and unweighted Dynamical-Discretized Reward Field(DDRF)isproposedasamajorcontributiontomotorcyclebehaviormodeling in mixed traf?c conditions. Other contributions of this research are: integration of a motorcycle trajectory maneuver model in the state transition function, derivation ofaprobabilityfunction,areaofawareness(AoA)andit’ssectorisationtoperceive vehicles inside the AoA, which is used in the determination of actions. A simulation was conducted to test the performance of the proposed model by comparingthedatafromthesimulationwithactualdata. Inthisstudy,weused100 data on motorcycle maneuvering, which consisted of two different scenarios, which are 50 data of motorcycle maneuvering to avoid other motorcycles and 50 data of motorcycle maneuvering to avoid cars. The environment setting of the simulation was adjusted to the actual conditions. The parameters used to verify the model were: depth of reward level, discounted factor or gamma factor, AoA radius, and a weighting function for the reward model. The performance of the proposed model was measured using root mean square error (RMSE). Ingeneral,basedonacomparisonofthesimulationresultswiththeactualdata,the proposed method can properly model the behavior of motorcycles in heterogeneous traf?c with an RMSE value of around 0.74 meters. This result has better performancetwicecomparedwiththecar-followingmodel. Therewardfunctionproposed in this study performed better than the reward function in previous studies arround 4-6%. Apart from that, a number of speci?c conclusions can be drawn based on the simulation results: the RMSE of motorcycle maneuvering motorcycle is greater than motorized scenario against a motorbike is greater than motorcycle maneuveringmotorcycle,whichshowsthatmodelingmotorcyclemaneuveringmotorcycle is more dif?cult than mdeling maneuvering a car caused by motorcycle is highly dynamics behavior; the proposed method performed well at average reward for depth of levels 1 and 2, which shows that in vehicle movement, riders tend not to thinkaboutthepossibilityofmovingtoofarahead;theeffectofAoAontheproposed method shows that motorists tend to have a moderate AoA coverage while riding; the discount factors performed well at small range values; and adding a weighting functiontotherewardmodelledtobetterperformance,especiallyforHammingand Bartlett weighting. Furthermore, the results of this study provide further research opportunities to be implemented in the behavior of four-wheeled riders, as well as opportunitiestodevelopthefunctionsofAoA,MRDT,probabilitytoimprovemodel performance.