Personalized optimal bicycle trip planning based on Q-learning algorithm

Yun Chen, Wen Yan, Chunguo Li, Yongming Huang, Luxi Yang

Abstract: Traveling by bicycle has become a rising trend recently for its convenience and flexibility, which calls for considerate bicycle trip planning schemes. While research for traditional trip planning has focused on quantized quality of point-of-interest (POI) or correlations among POIs, problems appear for distinct influential factors in bicycle trips and being unable to plan in a foreseeable stage with satisfying various demands of cyclists. In this paper, to alleviate the deficiencies of conventional approaches that merely concentrating on temporary interests and fully depending on greedy algorithm, the active Q-learning algorithm derived from reinforcement learning (RL) is adopted for Q-value iteration for planning overall optimal bicycle trips. To further meet personal improvised demands such as containing some specific places in the trip, Tailored Trip is provided and a dynamic and flexible place inserting algorithm is proposed to automatically tweak the trip and keep the planning optimum status. Experiments have been conducted to intuitively evaluate the performance of our schemes on two real-world datasets. The planning results clearly illustrate that the optimal node choosing policy is continuously reinforced in our schemes and the result for Tailored Trip highlights the guarantee of overall superiority after trip tweaking.