Autonomous Driving Research

Report on Current Developments in Autonomous Driving Research

General Direction of the Field

The field of autonomous driving is witnessing a significant shift towards more intelligent, collaborative, and privacy-conscious systems. Recent developments highlight the integration of advanced technologies such as Large Language Models (LLMs), Vision Language Models (VLMs), and federated learning to enhance system capabilities, improve safety, and address privacy concerns. The focus is on creating systems that are not only efficient and real-time but also adaptable to diverse and dynamic environments.

  1. Edge-Cloud Collaboration: There is a growing trend towards edge-cloud collaborative systems that leverage LLMs to manage and process data more efficiently. These systems use selective data upload mechanisms and advanced algorithms to optimize resource use and reduce latency, ensuring robust performance in real-world conditions.

  2. Inter-Vehicle Communication: The field is exploring novel methods for vehicles to share and utilize each other's field-of-view through onboard LLMs, enhancing environmental awareness and safety, especially in occluded scenarios.

  3. Pedestrian and Cyclist Interaction: Research is intensifying on understanding and improving interactions between autonomous vehicles and vulnerable road users like pedestrians and cyclists. This includes using eye-tracking, virtual reality, and electrodermal activity monitoring to gauge perception and stress levels.

  4. Privacy-Preserved Monitoring: There is a significant push towards developing privacy-preserving monitoring systems that use computer vision and LLMs to generate textual reports of pedestrian activity, thereby eliminating the need for video footage and addressing privacy concerns.

  5. Data Standardization and Integration: The use of LLMs for automated data standardization is emerging as a key area, enabling seamless integration and compatibility of heterogeneous sensor data from various sources, thereby improving positioning accuracy and system efficiency.

  6. Social Norms and Coordination: Exploring how LLMs can understand and model social norms in autonomous driving scenarios is a novel approach, aiming to foster coordination among autonomous vehicles without hard-coded rules.

Noteworthy Developments

  • Edge-Cloud Collaborative Motion Planning: Introducing EC-Drive, a system that optimizes communication resource use and reduces inference latency by selectively uploading critical data to the cloud for processing by GPT-4.
  • Video-to-Text Pedestrian Monitoring (VTPM): A framework that generates real-time textual reports of pedestrian activity while preserving privacy, reducing memory usage significantly.
  • Automated Data Standardization using LLMs: A feasibility study demonstrating the potential of LLMs in overcoming sensor data integration complexities for more precise IoT navigation solutions.

These developments not only advance the technical capabilities of autonomous driving systems but also address critical issues such as privacy, safety, and adaptability, marking significant progress in the field.

Sources

Edge-Cloud Collaborative Motion Planning for Autonomous Driving with Large Language Models

Understanding cyclists' perception of driverless vehicles through eye-tracking and interviews

V-RoAst: A New Dataset for Visual Road Assessment

Tapping in a Remote Vehicle's onboard LLM to Complement the Ego Vehicle's Field-of-View

Decoding Pedestrian Stress on Urban Streets using Electrodermal Activity Monitoring in Virtual Immersive Reality

Video-to-Text Pedestrian Monitoring (VTPM): Leveraging Computer Vision and Large Language Models for Privacy-Preserve Pedestrian Activity Monitoring at Intersections

Exploring the Feasibility of Automated Data Standardization using Large Language Models for Seamless Positioning

Can LLMs Understand Social Norms in Autonomous Driving Games?

Enhancing Vehicle Environmental Awareness via Federated Learning and Automatic Labeling

Built with on top of