Intelligent Amphibious Ground-Aerial Vehicles: State of the Art Technology for Future Transportation

Xinyu Zhang, Jiangeng Huang, Yuanhao Huang, Kangyao Huang, Lei Yang, Yan Han, Li Wang, Huaping Liu, Jianxi Luo, Jun Li

Introduction

While Urban Air Mobility (UAM) concepts like air taxis are gaining traction, they are often restricted to "spatial corridors"—essentially fixed bus routes in the sky. In contrast, amphibious ground-aerial vehicles offer far more flexibility by combining roadway driving with flight. This paper provides a comprehensive review of these vehicles, moving beyond the "flying car" novelty to explore the technical architecture required for a functional 3D urban transportation system. The authors argue that the true bottleneck isn't just aerodynamics or battery life, but the intelligent systems—specifically AI-driven perception, decentralized path planning, and adaptive control—needed to manage the complexity of land-air transitions.

The Intelligent System Framework

"Fly-driving" introduces a massive jump in complexity. Traditional automotive sensors aren't designed for the speeds of flight, and flight controllers often struggle with the "ground effect" (aerodynamic interference) during take-off and landing in tight urban spaces. The paper proposes a four-pillar framework for the intelligent system required for these vehicles:

  1. Intelligent Connected Systems: Moving beyond 4G/5G, which struggles at altitude, the role of Low Earth Orbit (LEO) satellites and High-Altitude Platforms (HAPs) is critical to ensure low-latency Vehicle-to-Infrastructure (V2I) communication.
  2. Near-Ground Perception: LiDAR and vision systems must evolve to handle "non-cooperative objects" (like birds or unregulated drones) and perform real-time slope detection for emergency landings.
  3. Decision-Making & Planning: 3D path-planning algorithms (A, Particle Swarm Optimization, and Genetic Algorithms) and decentralized multi-agent frameworks like PRIMAL2* are necessary to prevent mid-air collisions.
  4. Intelligent Control: This focuses on the "modality switch"—the transition between driving and flying—and using adaptive control to compensate for the turbulent airflow encountered near the ground.
The PAL-V Liberty prototype demonstrating the ground-driving mode of an amphibious vehicle.

AI-Driven Perception and Control

One of the most significant challenges is the aerodynamic interference caused by flying close to the ground. AI controllers must dynamically model this "ground effect" to prevent crashes during the most critical phases of flight. Furthermore, while a car might drive at 100 km/h, these vehicles fly at 180–360 km/h. At these speeds, motion blur becomes a fatal flaw for standard computer vision. The paper highlights CNN-based motion deblurring and Generative Adversarial Networks (GANs) as essential tools for high-speed perception.

For a city with thousands of flying vehicles, a central "air traffic control" would likely collapse under the data load. The authors advocate for decentralized, multi-agent path planning where vehicles "negotiate" their routes in real-time. This distinguishes amphibious vehicles from UAM; while UAM is a public transit model (fixed hubs), amphibious vehicles are the 3D equivalent of the private car, requiring much higher levels of individual autonomy.

Current Prototypes and Performance

The paper summarizes the specifications of nearly 20 modern prototypes, including KleinVision’s AirCar, PAL-V Liberty, and Terrafugia Transition.

  • Range: Fixed-wing hybrids currently dominate endurance, with some reaching up to 1000 km.
  • Payload: Most current models are limited to 200–400 kg, highlighting the ongoing struggle with power density.
  • Safety: The authors highlight the "Safe2Ditch" architecture, an AI-driven emergency system that identifies the safest possible landing spot (e.g., avoiding schools or busy intersections) when a mid-air failure occurs.
The KleinVision AirCar in flight, representing the current state of fixed-wing amphibious prototypes.

Limitations and Future Outlook

Despite rapid progress, several barriers remain. While Deep Reinforcement Learning (DRL) shows promise for 3D path planning in simulations, the real-world unpredictability remains a major barrier. Additionally, most current vehicles must come to a complete stop to fold or unfold wings; future research needs to focus on dynamic switching—transitioning from driving to flying without stopping.

Beyond AI, the physical limitations of battery energy density and the social "noise pollution" of rotors in residential areas remain unsolved hurdles. Finally, the technology is outpacing the law; 3D traffic regulations and "airspace sharing" rules are still in their infancy, requiring a robust legal framework to match the technical advancements.

Summary updated

BibTeX
@article{zhang2023intelligent,
  author = {Xinyu Zhang and Jiangeng Huang and Yuanhao Huang and Kangyao Huang and Lei Yang and Yan Han and Li Wang and Huaping Liu and Jianxi Luo and Jun Li},
  title = {{Intelligent Amphibious Ground-Aerial Vehicles: State of the Art Technology for Future Transportation}},
  journal = {IEEE Trans. Intell. Veh.},
  year = {2023},
  doi = {10.1109/tiv.2022.3193418},
}
Figure