A Method For Safely Scheduling Computing Task Offloads For Autonomous Vehicles

Brief Description

EnergyShield is a pioneering framework designed to optimize energy consumption through safe, intelligent offloading of deep neural network computations for autonomous vehicles.

Full Description

EnergyShield introduces a groundbreaking approach to manage the computational demands of neural network controllers in autonomous vehicles. By offloading tasks to edge computing resources in a provably safe manner, it ensures that the vehicle's safety and performance are not compromised. This is achieved through a novel safety monitor that acts as a "shield," allowing for energy-efficient computations without sacrificing the reliability required by mission-critical systems.

Suggested uses

• Autonomous driving systems requiring real-time, safe, and energy-efficient computation. 
• Embedded systems in autonomous vehicles, unmanned aerial vehicles (UAVs), drones, and robotics with limited computational resources. 
• Edge computing platforms seeking to support mission-critical applications with stringent safety requirements.

Advantages

• Ensures formal safety properties of autonomous systems while enabling low-power offloading optimizations. 
• Designed with a comprehensive perspective, making it adaptable and effective. 
• Introduces a novel use of safety filters as runtime monitors to guide offloading decisions. 
• Achieves energy efficiency gains without compromising on safety guarantees. 
• Offers a scalable and generic methodology applicable across various controllers, safety functions, and autonomous systems.

Patent Status