Environmental-Driven Approach towards Level 5 Self-Driving.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Science

Background:

• Advancements in technology are driving the development of fully autonomous vehicles (level 5).
• Achieving level 5 autonomous driving requires significant research beyond level 4, particularly for safe operation in complex environments.
• Real-time performance is critical, as sensor systems and machine learning algorithms in autonomous cars must meet strict end-to-end delay deadlines.

Purpose of the Study:

• To propose a novel framework for autonomous cars to ensure end-to-end delay deadlines.
• To address the computational intensity of machine learning modules in autonomous driving systems.
• To maintain driving accuracy while meeting performance requirements.

Main Methods:

• An environment-driven approach was developed, distinguishing between uncontrollable environmental factors and controllable internal factors (e.g., sensing frequency, image resolution, prediction rate, car speed).
• An admission control module was designed to regulate internal parameters like image resolution and detection period.
• The framework determines the acceptability of parameters for meeting end-to-end deadlines based on the current environmental scenario.

Main Results:

• The proposed framework successfully manages internal parameters to meet end-to-end deadlines.
• The system demonstrated its ability to maintain autonomous driving accuracy under various conditions.
• Verification was performed using both a remote-controlled (RC) car and a simulation environment.

Conclusions:

• The environment-driven framework offers a viable solution for ensuring timely performance in autonomous vehicles.
• Controlling internal parameters based on environmental context is key to achieving reliable level 5 autonomous driving.
• The approach provides a method for balancing computational load and driving accuracy in complex autonomous systems.