FusionHub

FusionHub is Xikaku’s real-time sensor fusion engine. It ingests raw data from IMUs, GNSS/RTK receivers, optical trackers, wheel odometry, CAN buses, and VR headsets, fuses them through production-grade Unscented Kalman Filters, and delivers a single source of truth — pose, motion, and navigation data you can act on. The same engine drives autonomous vehicles, AGVs, mocap stages, AR/VR headsets, and in-flight pilot training rigs.

FusionHub runs as a native desktop application or headless on the edge, on Windows and Linux, and is embeddable in other software via C FFI.

Connecting AI to the physical world. Modern AI systems are only as useful as the world model they can act on. FusionHub is the bridge between raw sensor streams and AI — it exposes fused pose, motion, and navigation state over a built-in Model Context Protocol (MCP) server, so language models, agents, and vision systems can query live ground truth, reason about it, and feed decisions back into the pipeline. The first-party AI Monitor extension, powered by Claude, watches running pipelines for anomalies and drift, turning sensor-fusion telemetry into actionable insight without custom glue code.

A visual, node-graph approach to sensor fusion

Every FusionHub pipeline is a graph of three node types: sources that read sensor data, filters that fuse or transform it, and sinks that hand the fused result to downstream software. Nodes are wired together in a browser-based editor and configured with JSON, with hot reload so changes apply without restarts.

Because every node exposes typed inputs and outputs, the editor validates connections live — you can’t plug a GNSS stream into a port that expects quaternions. Property panels are generated automatically from each node’s metadata, and a built-in oscilloscope lets you inspect live data on any link.

Supported sensor inputs

FusionHub ships with first-party integrations for a wide range of industry-standard hardware and protocols:

• IMUs — LP-Research LPMS family (B2, U3, AL3, IG1, IG1P, IG1W, NAV3, CURS3, INC1), HID, Xreal One Pro built-in IMU, PX4 DDS IMU streams
• Optical tracking — ART DTrack, Vicon, OptiTrack, Antilatency, LPVISION, Isaac ROS Visual Odometry
• GNSS & RTK — NMEA receivers, RTCM correction streams
• Automotive & odometry — PEAK and Vector CAN (with DBC presets), ACT2535 velocity meters, ELM327 OBD-II
• VR / AR headsets — ALVR (Meta Quest, VIVE Focus Vision, Apple Vision Pro), OpenVR / SteamVR
• Generic I/O — MQTT, file playback (for replaying recorded sessions), UART, TCP

Fusion filters

FusionHub’s filters implement the actual sensor fusion mathematics — every algorithm is tuned for real-world noise, dropouts, and latency:

• IMU + Optical Fusion — 6-DoF pose estimation with automatic intercalibration between the IMU and optical frame
• GNSS + IMU Fusion — 20-state Unscented Kalman Filter with online bias and antenna-offset estimation; handles GNSS dropouts gracefully with IMU-only prediction
• Vehicular Fusion — 2-D dead-reckoning from wheel odometry + IMU
• Full Vehicle Fusion — multi-stage pipeline combining odometry + IMU and GNSS + IMU for automotive and AGV workloads
• Differential IMU — aligns a pair of IMUs so the difference can be used as a reference (e.g. vehicle-mounted vs head-mounted)
• Prediction — pose extrapolation for latency compensation
• Intercalibration — automatic alignment of sensor frames without manual surveying

Outputs and integrations

Sinks deliver the fused result to downstream software in whatever format it expects:

• Standards — NMEA (serial and TCP), VRPN, DTrack protocol proxy, ROS 2 publisher, PX4 DDS GPS source
• Messaging — MQTT publisher, WebSocket server
• VR / AR — LPVIZ SteamVR display driver (Xreal One Pro, Xreal Air 2 Ultra), ALVR bridge (Meta Quest, Focus Vision, Vision Pro)
• Recording & debugging — file logger, echo sink, live oscilloscope
• Emulation — ACT2535 vehicle-velocity emulation for test rigs

Built-in web UI

FusionHub’s web UI runs on localhost at port 19359 and is embedded in the single binary — no separate install. It provides a live dashboard with per-node status tiles and I/O counters, a full-featured node editor, log viewer, license manager (server / file / USB token), cloud workspace sync, and an in-app updater.

The UI also hosts a set of extension pages that activate automatically based on the configured nodes: a 3-D pose viewer, a global-coordinate map view, a calibration tool, an AI Monitor for anomaly detection, a Rhai script editor, data playback controls, LPMS sensor configuration (CAN/CANOpen/UART setup, filter tuning, firmware updates), and an ALVR control panel.

Complex configurations at a glance

FusionHub scales from a single IMU on a workbench to multi-sensor production rigs. The example below shows a full LPVIZ + LPPOS vehicle-tracking setup, fusing GNSS, a vehicle-motion feed, and an ACT2535 velocity meter in one filter chain while a head-mounted IMU, a reference IMU, and DTrack optical tracking feed another — simultaneously driving a SteamVR display driver, a 3-D view, a ROS 2 publisher, and a map sink.

Common deployments

• LPVR-DUO with Xreal One Pro / Xreal Air 2 Ultra — drives an LPVIZ SteamVR driver from head-IMU + optical tracking
• LPVR-CAD with Apple Vision Pro — inside-out fusion over ALVR for design review
• ALVR with Meta Quest or VIVE Focus Vision — wireless VR streaming with Xikaku-tracked head pose
• Full LPVIZ + LPPOS — in-car mixed reality with integrated vehicle tracking

Detailed setup guides for each live in the FusionHub RS docs.

Platform and licensing

• Operating systems — Windows and Linux, native single binary
• Integrations — C FFI for embedding in native apps; MCP server interface for AI-assisted monitoring and control
• Extensibility — nodes self-register via a global registry, so adding new sources, filters, or sinks does not require changes to the core
• Licensing — feature-gated (IMU + optical fusion, vehicular fusion, differential IMU, etc.); licenses issued per feature and activated via server, file, or USB token