On Social Interactions of Merging Behaviors at Highway On-Ramps in Congested Traffic

Abstract

Merging at highway on-ramps while interacting with other human-driven vehicles is challenging for autonomous vehicles (AVs). An efficient route to this challenge requires exploring and then exploiting knowledge of the interaction process from demonstrations by humans. However, it is unclear what information (or the environment states) is utilized by the human driver to guide their behavior over the whole merging process. This paper provides quantitative analysis and evaluation of the merging behavior at highway on-ramps with congested traffic in a volume of time and space. Two types of social interaction scenarios are considered based on the social preferences of surrounding vehicles: courteous and rude. The significant levels of environment states for characterizing the interactive merging process are empirically analyzed based on the real-world INTER-ACTION dataset. Experimental results reveal two fundamental mechanisms in the merging process: 1) Human driver selects different states to make sequential decisions at different moments of task execution and 2) the social preference of surrounding vehicles has an impact on variable selection for making decisions. It implies that for autonomous driving, efficient decision-making design should filter out irrelevant information while considering the social preference of the surrounding vehicles, to reach a comparable human-level performance. These essential findings shed light on developing new decision-making approaches for AVs.

Uncovering Interpretable Internal States of Merging Tasks at Highway On-Ramps for Autonomous Driving Decision-Making

“…The data processing in this section is based on our previous research [25]. The definition of vehicles (i.e., ego vehicle, lead and lag vehicle), merge critical moments (i.e., start moment t s , middle moment t m , and end moment t e ), and social preference (rude or courteous) can refer to [25].…”

“…The data processing in this section is based on our previous research [25]. The definition of vehicles (i.e., ego vehicle, lead and lag vehicle), merge critical moments (i.e., start moment t s , middle moment t m , and end moment t e ), and social preference (rude or courteous) can refer to [25]. The sequential data during the whole process with courteous merging behavior between t s and t e are extracted and the merging event amounts to 789.…”

“…The variable selection in existing works usually relies on researchers' experience and on-board sensors. Our previous research [25] reveals that the critical variables change over the merging stages, and redundant variables should be removed as noises. Only proper variable selection can be conducive to the inference and learning of internal states and improve the decision-making performance.…”

“…Selection of appropriate input variables can eliminate the interference of redundant variables and maximize the performance of the HMM-GMR framework. Our previous research in [25] reveals that the most critical variables of making decisions when merging into highway are ranked as v ego y , v ego x , ∆v lead x , ∆x lag , ∆v lag…”

“…This paper provides a probabilistically learning approach to automatically extract the internal states of the multi-vehicle interactive process (rather than of a single vehicle's behavior) which can guide the ego vehicle to make an appropriate decision. Based on the conclusion of our previous research in [25], we here developed a probabilistic approach (i.e., HMM-GMR) to learn and reproduce the internal dynamics of merge tasks at highway on-ramps. The HMM-GMR framework combines HMM with Gaussian mixture regression (GMR) [26] and can leverage the temporal information into dynamic processes.…”