Eco-driving Accounting for Interactive Cut-in Vehicles

TL;DR

The paper addresses eco-driving for automated vehicles operating in traffic with potential cut-in neighbors. It introduces a leader-follower game-theoretic model to capture interactive cut-in behaviors and an MPC-based eco-driving controller that uses predictions of cut-in and preceding-vehicle motions to plan energy-efficient actions. A Bayesian estimation framework updates the cut-in vehicle's intended role, and predicted motions are fused to produce robust, probabilistic forecasts for safe yet efficient driving. Across simulations, the approach yields substantial energy savings compared with baselines that ignore cut-ins, demonstrating improved performance when handling interactive cut-in scenarios attacked by leader or follower strategies.

Abstract

Automated vehicles can gather information about surrounding traffic and plan safe and energy-efficient driving behavior, which is known as eco-driving. Conventional eco-driving designs only consider preceding vehicles in the same lane as the ego vehicle. In heavy traffic, however, vehicles in adjacent lanes may cut into the ego vehicle's lane, influencing the ego vehicle's eco-driving behavior and compromising the energy-saving performance. Therefore, in this paper, we propose an eco-driving design that accounts for neighbor vehicles that have cut-in intentions. Specifically, we integrate a leader-follower game to predict the interaction between the ego and the cut-in vehicles and a model-predictive controller for planning energy-efficient behavior for the automated ego vehicle. We show that the leader-follower game model can reasonably represent the interactive motion between the ego vehicle and the cut-in vehicle. More importantly, we show that the proposed design can predict and react to neighbor vehicles' cut-in behaviors properly, leading to improved energy efficiency in cut-in scenarios compared to baseline designs that consider preceding vehicles only.

Figures (7)

• Figure 1: Top view of the scenario studied in this work. The automated ego vehicle (blue vehicle 0) is approaching slow traffic ahead (with grey vehicle 2 at the tail). In the lane on the left, there is also slow traffic building up (with grey vehicle 3 at the tail) and a target vehicle (red vehicle 1) is approaching this traffic and has the intention to cut into the ego vehicle's lane.
• Figure 2: Nonlinear functions in the vehicle dynamics. (a) Saturation function \ref{['eqn:sat_def']}. (b) Acceleration limits \ref{['eqn:umax_def']}.
• Figure 3: Nonlinear functions in the optimal velocity model (OVM). (a) Range policy \ref{['eqn:range_policy']}. (b) Speed policy \ref{['eqn:speed_policy']}.
• Figure 4: The time profile of the ego vehicle with different controllers and scenarios (a) longitudinal speed (b) distance headway (c) longitudinal acceleration. In all panels, the blue solid curves correspond to the case when the ego is using OVM controller \ref{['eqn:OVM']} and no vehicle cut-in from the adjacent lane; the red dashed curves correspond to the case when the ego is using baseline eco-driving controller with no vehicle cut-in from the adjacent lane; the green dashed-dotted curves correspond to the case when the ego is using the eco-driving controller in Algorithm \ref{['alg:eco_driving']} with a vehicle cut-in from behind from the adjacent lane, and the first 4 seconds top view of this scenario is shown in Fig. \ref{['fig:cut_in_from_behind_top_view']}.
• Figure 5: Top view of the scenario when the cut-in vehicle cut from the behind of the ego vehicle, with the screenshot corresponding to every second up to the 4 seconds, which covers the whole cut-in procedure. In all screenshots, the blue vehicle is the ego, the red vehicle is the cut-in vehicle and the grey vehicles are other vehicles traveling in the slow traffic ahead. The ego (blue vehicle) motion corresponds to the first 4 seconds of the green dashed-dotted profiles in Fig. \ref{['fig:no_cutin_vs_cut_in_from_behind']}.