V2X-Communication-Aided Autonomous Driving: System Design and Experimental Validation

Chanyoung Jung, Daegyu Lee, Seungwook Lee, David Hyunchul ShimSensors 2020 · DOI

TL;DR

This paper presents a V2X-enabled autonomous driving system that enhances perception and planning by fusing on-board sensor data with vehicle-to-everything (V2X) communication. The system was successfully validated in real-world scenarios at the Hyundai Autonomous Vehicle Competition, demonstrating its capability to handle complex driving situations beyond the limitations of traditional sensors.

What Problem Does This Paper Address?

Autonomous driving systems rely heavily on sensors like cameras, LiDAR, and radar for perception. However, these sensors have a limited line-of-sight (LOS), making it difficult to perceive hazards at intersections or around blind corners. This limitation significantly impacts the robustness and safety of planning and control subsystems. While Vehicle-to-Everything (V2X) communication offers a way to overcome this, existing research often focuses on specific V2X applications (like platooning) or relies solely on simulations. This paper addresses the need for a comprehensive V2X-aided autonomous driving system that integrates V2X communication into perception, planning, and control for highly automated vehicles and validates its performance in realistic, real-world conditions.

Methods

The proposed system is built around three core subsystems:

  • Beyond Line-of-Sight (BLOS) Perception: This subsystem fuses data from on-board sensors (local perception) with information received via V2X (communication perception). This fusion aims to extend the perception range, especially in occlusion-prone environments. A data fusion algorithm is employed to identify dynamic obstacles (vehicles, pedestrians) by comparing and correlating data from both sources.
  • Extended Planning: This subsystem incorporates real-time traffic information obtained through V2X. It includes:
  • Global Path Planner: Uses a lightweight road network model and considers V2I traffic information (like construction sites) to plan a cost-minimized route, detouring around known obstacles.
  • Local Path Planner: Generates collision-free paths considering vehicle dynamics and constraints, using a combination of road-model-based and graph-based (hybrid A-star) approaches.
  • Behavior Planner: Manages the vehicle's state through a finite state machine (FSM) with 11 states, handling various driving scenarios like traffic flow, potential hazards, and intersections.
  • Velocity Planner: Sets appropriate speeds based on traffic signals, speed limits, and the planned path, ensuring smooth acceleration and deceleration.
  • Control: This subsystem implements low-level longitudinal and lateral control.
  • Longitudinal Control: Uses a cascade feedback structure with an outer speed loop and an inner pedal loop to control acceleration and braking.
  • Lateral Control: Employs a modified Stanley method with a "look-ahead point" to improve stability and reduce overshoot in path following.

Datasets

The paper does not explicitly mention training datasets. The system was evaluated using real-world data collected during the 2019 Hyundai Autonomous Vehicle Competition held in K-city, a dedicated test bed equipped with V2X infrastructure. The competition involved various missions simulating real-road scenarios.

Key Insights and Ideas

  • V2X as a Complementary Sensor: The core idea is that V2X isn't a replacement for on-board sensors but a powerful complement, especially for overcoming LOS limitations. The BLOS perception system effectively demonstrates this by fusing data to gain a more comprehensive environmental understanding.
  • Integrated System for High Automation: The paper emphasizes the need for a tightly integrated system for highly automated vehicles (Level 4+), where perception, planning, and control must work seamlessly. The proposed three-subsystem architecture addresses this.
  • Real-World Validation is Crucial: Relying solely on simulations is insufficient for autonomous driving. The successful participation and performance in the Hyundai AVC competition highlight the importance of testing in realistic, dynamic environments with actual V2X infrastructure.
  • Dynamic Path Replanning: The system's ability to replan global paths in response to V2I information (like construction sites) and generate collision-free local paths using hybrid A-star when road-model-based candidates fail is a key strength.
  • Behavioral FSM for Robustness: The 11-state FSM for behavior planning allows the system to adapt to a wide range of urban driving scenarios, preventing it from getting stuck and ensuring smooth transitions between different driving modes.

Key Results

The proposed V2X-aided autonomous driving system, integrated into a full-scale Hyundai i30 vehicle named "Eurecar," achieved third place in the 2019 Hyundai Autonomous Vehicle Competition. The system successfully completed all assigned missions, including:

  • Emergency Braking: Stopped within 3 meters of a jaywalker, demonstrating effective pedestrian detection and rapid braking.
  • Detouring Construction Sites: Successfully replanned its global path to avoid construction zones based on V2I information.
  • Complying with Traffic Signals: Navigated intersections by adhering to traffic light status received via V2I, achieving fast lap times.
  • Collision Avoidance: Successfully navigated an accident scene with multiple lanes and obstacles, utilizing both road-model-based and graph-based path planning.
  • Yielding to Emergency Vehicles: Detected an approaching emergency vehicle via V2V and safely changed lanes to yield.

Eurecar also recorded the fastest lap time (3 min 30 s) among 12 teams in the final round, running at a top speed of approximately 70 km/h.

Limitations and Future Work

The authors acknowledge several limitations and suggest future research directions:

  • Non-Signalized Intersections: The current system warns of imminent collisions at non-signalized intersections and decelerates, rather than proactively perceiving other vehicles' intentions. Future work should address this by improving perception and prediction in such scenarios.
  • Wider Variety of Traffic Entities: The system needs further testing and improvement to handle a broader range of traffic participants, such as bicycles and "cheating" vehicles (vehicles that violate traffic rules).
  • Expanding Scope: The authors plan to expand the research scope to further improve the system and contribute to making fully autonomous vehicles a reality. This could involve more complex urban scenarios and integration with advanced AI techniques.

Summary updated

BibTeX
@article{jung2020v,
  author = {Chanyoung Jung and Daegyu Lee and Seungwook Lee and David Hyunchul Shim},
  title = {{V2X-Communication-Aided Autonomous Driving: System Design and Experimental Validation}},
  journal = {Sensors},
  year = {2020},
  doi = {10.3390/s20102903},
}
Figure