Towards Energy-Efficiency by Navigating the Trilemma of Energy, Latency, and Accuracy

TL;DR

This work tackles the energy efficiency challenges in XR scene reconstruction by addressing the trilemma of energy, latency, and accuracy in TSDF Fusion. It introduces three optimization classes—algorithm, execution, and data—that collectively define a large design space of 72 configurations and reveal a Pareto frontier achievable only through co-optimization and consideration of downstream consumer needs. Empirical results on an embedded platform show energy reductions up to $60\times$, with latency and accuracy trade-offs depending on constraints, and a ScanNet case achieving around $25\times$ energy savings with modest latency impact and negligible quality loss. The findings highlight the necessity of system-wide optimization for XR perception tasks and provide actionable guidance for deploying energy-efficient designs on mobile XR devices and beyond.

Abstract

Extended Reality (XR) enables immersive experiences through untethered headsets but suffers from stringent battery and resource constraints. Energy-efficient design is crucial to ensure both longevity and high performance in XR devices. However, latency and accuracy are often prioritized over energy, leading to a gap in achieving energy efficiency. This paper examines scene reconstruction, a key building block for immersive XR experiences, and demonstrates how energy efficiency can be achieved by navigating the trilemma of energy, latency, and accuracy. We explore three classes of energy-oriented optimizations, covering the algorithm, execution, and data, that reveal a broad design space through configurable parameters. Our resulting 72 designs expose a wide range of latency and energy trade-offs, with a smaller range of accuracy loss. We identify a Pareto-optimal curve and show that the designs on the curve are achievable only through synergistic co-optimization of all three optimization classes and by considering the latency and accuracy needs of downstream scene reconstruction consumers. Our analysis covering various use cases and measurements on an embedded class system shows that, relative to the baseline, our designs offer energy benefits of up to 60X with potential latency range of 4X slowdown to 2X speedup. Detailed exploration of a use case across representative data sequences from ScanNet showed about 25X energy savings with 1.5X latency reduction and negligible reconstruction quality loss.

Figures (9)

• Figure 1: We study three classes of optimizations that create a large design space through configurable parameters, where the designs provide different trade-offs in the dimensions of energy, latency, and accuracy. Co-optimization across all classes leads to significant energy savings within latency and accuracy constraints, illustrated on the right for a case that additionally provides latency benefits with negligible accuracy loss.
• Figure 2: The breakdown of processed voxel status shows that only about 40% of voxels are fused (V. Fused). Our design filters out non-contributing voxels while preserving the critical ones nearly intact.
• Figure 3: Our design decides eligible voxels (in purple) from valid volumes (in pink) by testing 9 critical positions (5 in 2D). A center point outside the truncation band finds 1 or 2 sub-volumes (purple in Volume A), while a center point within the band finds all voxels eligible for fusion (Volume B).
• Figure 4: Energy and latency as functions of the number of threads. The sweet spot (4) in the energy-latency trade-off differs from the default (8), indicating that parallelization must be carefully implemented to balance overheads and benefits.
• Figure 5: Energy and latency as functions of CPU frequency. Energy and latency change in opposite directions and at different rates as frequency changes, motivating careful design for an energy efficient sweet spot in the energy-latency trade-off.