Ada: A Distributed, Power-Aware, Real-Time Scene Provider for XR.

Area of Science:

• Computer Science
• Extended Reality (XR) Systems
• Spatial Computing

Background:

• Real-time scene provisioning is crucial for spatial computing in XR.
• Existing solutions face challenges in balancing latency, power, and scene fidelity on XR devices.
• Current systems are often closed or application-specific.

Purpose of the Study:

• To introduce Ada, the first open, distributed, power-aware, and application-agnostic real-time scene provisioning system.
• To demonstrate Ada's ability to provide high-fidelity scenes with stable performance and low power consumption.
• To evaluate Ada's performance against prior work and showcase the benefits of its innovations.

Main Methods:

• Developed Ada with computation offloading, algorithmic, and system innovations.
• Configured an on-device, CPU-based variant (AdaLocal-CPU) for direct comparison with prior work.
• Implemented a distributed GPU-accelerated version of Ada.

Main Results:

• AdaLocal-CPU achieved up to 6.8x lower scene request latency and higher fidelity than prior work.
• The distributed GPU-accelerated Ada reduced latency by an additional 2x.
• Ada lowered incremental power cost by 24% compared to the best on-device variant (AdaLocal-GPU).

Conclusions:

• Ada offers a flexible solution adaptable to diverse latency, power, fidelity, and bandwidth requirements.
• Ada successfully balances latency, power, and scene fidelity for XR applications.
• Ada represents a significant advancement in open, distributed, and power-aware real-time scene provisioning.