Human-Machine Shared Control Approach for the Takeover of CACC

Haoran Wang, Zhenning Li, Arno Eichberger, Jia Hu, Eichberger, Arno, Hu, Jia, Chen, Yongyu, Gao, Yongji

TL;DR

This paper introduces a novel shared-control framework to manage the risky "handover" moment when a human driver takes over from a Cooperative Adaptive Cruise Control (CACC) system. By modeling the interaction as a Stackelberg game—where the machine acts as a leader that "guides" the human follower—the system ensures a smooth transition, maintains platoon stability, and reduces upstream traffic disturbances by up to 60%.

What Problem Does This Paper Address?

Cooperative Adaptive Cruise Control (CACC) is a powerful technology that allows vehicles to follow each other with extremely small gaps (as low as 0.5 seconds) thanks to vehicle-to-vehicle (V2V) communication. While this doubles road capacity and saves fuel, it creates a "takeover paradox."

When a human needs to take control—for instance, to exit a freeway—they often feel unsafe being so close to the car ahead. This "unsafe feeling" leads to abrupt, hard braking. In a platoon of high-speed vehicles, one driver’s panic-braking can ripple backward, causing "string instability" (oscillations that grow larger as they move back) and potentially leading to multi-car collisions or massive traffic jams. Current systems usually treat takeover as a binary switch, but this paper argues for a gradual, shared-control "handshake."

Methods

The authors propose an indirect human-machine shared control approach based on three sophisticated pillars:

  1. Stackelberg Game Modeling: Instead of just predicting what the human will do, the machine acts as the "Leader" and the human as the "Follower." The machine plans its actions (acceleration/braking) knowing how the human is likely to react, effectively "nudging" the human toward safer, smoother behavior.
  2. Game-based Model Predictive Control (GMPC): The machine uses GMPC to optimize its commands over a rolling time horizon. It integrates a "human reaction function" directly into its own planning logic, ensuring that the final fused command is mathematically optimal for the whole system.
  3. Inverse Reinforcement Learning (IRL): To make the system personalized, the researchers used IRL to identify individual driving styles in real-time. By observing how a specific person drives, the system identifies their unique weighting for speed error vs. following gap, allowing the controller to adapt to aggressive or conservative drivers with 91.6% accuracy.

The final output is a Control Fusing Law that blends human and machine inputs, gradually shifting authority from 100% machine to 100% human over a set duration.

Datasets

The study was evaluated using a simulation environment built on Matlab.

  • Human Behavior Data: The researchers used real-time collected human driving data to train and validate the IRL-based driving style identification.
  • Scenario Testing: They tested the controller against three specific cases: constant speed cruising, sinusoidal speed oscillations (to test stability), and emergency hard braking.

No large-scale external public datasets were used; the focus was on the mathematical verification of the control laws within a high-fidelity simulation.

Key Insights and Ideas

  • Leading, Not Just Predicting: Traditional "prediction-planning" pipelines are reactive. By using a Stackelberg game, the machine becomes proactive. It understands that its own actions influence the human, allowing it to "dampen" the human's tendency to overreact.
  • The 40% Threshold: The researchers quantified the "Operational Design Domain" (ODD) for string stability. They found that if the human driver holds more than 40% of the control authority, the platoon typically becomes unstable (oscillations grow). This provides a concrete engineering limit for how fast authority can be handed over.
  • The Platoon Limit: String stability is significantly harder to maintain as platoons grow. The study suggests that for safe human-in-the-loop takeovers, platoons should be restricted to fewer than 6 vehicles.

Key Results

  • Traffic Impact: The shared control approach reduced the impact on upstream traffic (the "ripple effect" of braking) by up to 60% compared to direct human takeover.
  • Safety Gains: Perceived safety (measured by the reduction in acceleration "jerks" or oscillations) improved by 53.23%.
  • Actual Safety: In emergency braking scenarios, the system prevented collisions that occurred under pure human control by ensuring the minimum following gap stayed above the safety threshold.
  • Efficiency: The distributed controller is computationally light, with an average computing time of just 0.05 seconds, making it viable for real-time hardware implementation.

Limitations and Future Work

The authors acknowledge a few key areas for further exploration:

  • Automation-Initiated Focus: This paper primarily looks at transitions initiated by the system (e.g., the car telling the human to take over). Driver-initiated transitions (where a human suddenly grabs control) involve different psychological factors and "voluntary overrides" that require more study.
  • Communication Delays: The current model assumes near-instantaneous V2V communication. Future iterations need to account for network latency and potential packet loss in the V2V link.
  • Complex Maneuvers: The study focuses on longitudinal control (speed and distance). Integrating lateral control (steering) for complex maneuvers like lane changes during takeover is a logical next step.

Summary updated

Abstract

Cooperative Adaptive Cruise Control (CACC) often requires human takeover for tasks such as exiting a freeway. Direct human takeover can pose significant risks, especially given the close-following strategy employed by CACC, which might cause drivers to feel unsafe and execute hard braking, potentially leading to collisions. This research aims to develop a CACC takeover controller that ensures a smooth transition from automated to human control. The proposed CACC takeover maneuver employs an indirect human-machine shared control approach, modeled as a Stackelberg competition where the machine acts as the leader and the human as the follower. The machine guides the human to respond in a manner that aligns with the machine's expectations, aiding in maintaining following stability. Additionally, the human reaction function is integrated into the machine's predictive control system, moving beyond a simple "prediction-planning" pipeline to enhance planning optimality. The controller has been verified to i) enable a smooth takeover maneuver of CACC; ii) ensure string stability in the condition that the platoon has less than 6 CAVs and human control authority is less than 40%; iii) enhance both perceived and actual safety through machine interventions; and iv) reduce the impact on upstream traffic by up to 60%.
BibTeX
@article{wang2025human,
  author = {Haoran Wang and Zhenning Li and Arno Eichberger and Jia Hu and Eichberger, Arno and Hu, Jia and Chen, Yongyu and Gao, Yongji},
  title = {{Human-Machine Shared Control Approach for the Takeover of CACC}},
  journal = {IEEE Trans. ITS},
  year = {2025},
  doi = {10.1109/tits.2025.3580756},
}
Figure