AR Glasses and Robots Can Now Track Motion on Any Chip
Cheaper, lighter headsets and robots — without buying expensive Nvidia hardware.
Researchers built a new way for robots and XR to figure out where they are and how they're moving — a job called visual-inertial odometry, which the paper says is key to state estimation. Their method, VkVIO, is described as the first cross-platform GPU-accelerated VIO method, using the vendor-agnostic Vulkan API instead of the CUDA that previous work in the literature had limited itself to, which had narrowed deployment to a single vendor. The authors report state-of-the-art accuracy with the causal estimates required for real-time operation, and they deployed VkVIO across a workstation, a laptop, and an extremely inexpensive single-board computer — outperforming CUDA-based systems on the same hardware. The paper says this enables possibilities for low-latency, low-power, and low-cost VIO in robotics and XR.
- VIO is the tech that lets robots and AR headsets know where they are by combining cameras with motion sensors — like your sense of balance and sight working together.
- Previous versions only ran on Nvidia chips (CUDA); VkVIO uses Vulkan, which works on virtually any graphics chip, so device makers aren't locked into one supplier.
- It ran in real time on an extremely cheap single-board computer, which points toward lighter, cooler, cheaper AR glasses, drones and robots.
Why It Matters
Cheaper, lighter AR glasses, drones and robots that run longer on a charge — no premium chip required.