CV - Job Market Paper  
 

Tikhonenko, Dmitrii  

Contact Information

Tel. +34 93 542 2692

dmitrii.tikhonenko@upf.edu

 

 

 

 

 

 

Research interests

Transportation Science. Operations Research. Game Theory. Optimization. Machine Learning.

 

Placement officer

Filippo Ippolito
[email protected]
 

References

Kalyan Talluri (Advisor)
[email protected]

Mihalis Markakis (Advisor)
[email protected]

Piotr Zwiernik
[email protected]


 

Research

"Managing Algoritm-Assisted Drivers" (Job Market Paper)
Why do traffic jams and stop-and-go waves happen? Is there a way to prevent them? Traffic theory models ascribe the fundamental cause of such phenomena to exogenous fluctuations originating from human driving behavior. But what if vehicles are controlled or assisted by algorithms? Would these just go away? I study a simple, yet very common forced-merging scenario: a two-lane road segment that has one lane blocked unexpectedly, say, due to an accident or construction. All cars on the blocked lane need to merge to the free lane in order to continue their routes. Motivated by the recent advent of algorithm-assisted driving, I assume that drivers are rational, self- interested agents, wishing to minimize their individual travel times, deciding (a) at what velocity to move; and (b) whether to merge to the free lane, given the opportunity (gap). Moving at higher velocities on the blocked lane reduces the travel time but also the chance of finding a large enough gap to merge, so blocked-lane drivers are trading off travel time vs. risk of not being able to merge. I analyze a dynamic programming formulation of the problem with a single merging driver, and characterize the optimal policy, which turns out to be a multi-threshold one with a surprising structure: in the presence of uncertainty regarding merging to the target lane, it may be optimal for a driver–in certain regions–to oscillate between high and low velocities while attempting to merge. Hence, the origin of traffic oscillations need not be purely “behavioral” but can arise endogenously as the outcome of optimization. I show how a central planner can set velocity limits for avoiding such oscillations, and test our policies via extensive discrete event simulations with multiple merging vehicles.