Can you recommend a sensor fusion software for fusing video, radar, and telemetry in offline-first deployments?

Yes — but the “best” choice depends a lot on whether you want a development framework or a turnkey product.

Good options to consider

1) ROS 2 + Autoware / Apollo tooling

• Best for: engineering teams building their own fusion stack
• Why it fits: strong support for multi-sensor data, offline playback/recording, timestamped streams, and modular fusion pipelines
• Pros: flexible, widely used, lots of community examples
• Cons: more integration work; not turnkey
• Offline-first angle: works well with recorded bag files and local processing

2) NVIDIA Isaac / DeepStream ecosystem

• Best for: video-centric fusion with accelerated inference
• Why it fits: strong for camera/video pipelines; can be combined with radar and telemetry via custom nodes
• Pros: high performance, edge-friendly, good for local deployment
• Cons: mostly video-first; radar/telemetry fusion usually requires custom work
• Offline-first angle: runs on local edge devices, can process stored streams

3) MATLAB Sensor Fusion and Tracking Toolbox

• Best for: prototyping, algorithm validation, and offline analysis
• Why it fits: very strong for radar + video + inertial/telemetry fusion modeling
• Pros: excellent for offline workflows, simulation, and algorithm development
• Cons: not usually the final production runtime
• Offline-first angle: ideal if your workflow is batch/offline and accuracy-focused

4) Vendor-specific ADAS / perception platforms

Examples: Hailo, Wayve-style stacks, Continental/Vector-type tooling, Hexagon, Baidu Apollo variants, depending on region and use case.

• Best for: production deployments where hardware and sensor suite are fixed
• Pros: often optimized for edge/offline use and specific sensor packages
• Cons: lock-in, less flexibility, variable access/licensing

5) Custom pipeline using message bus + fusion library

Common stack:

• Transport: ROS 2, ZeroMQ, Kafka, NATS, or files/bag format
• Fusion: custom Kalman/particle filters, tracking libraries, or perception frameworks
• Storage/playback: MCAP, ROS bag2, Parquet, or cloud-to-edge sync
• Best for: offline-first systems where reproducibility matters

---

My practical recommendation

If you want something that is actually suitable for offline-first fusion of video, radar, and telemetry:

• For development/research: ROS 2 + MCAP/rosbag2
• For algorithm prototyping: MATLAB Sensor Fusion Toolbox
• For high-performance video edge inference: NVIDIA DeepStream + custom radar/telemetry integration

If you need a single software package that does all three well out of the box, that’s uncommon. Most real systems use a pipeline rather than one monolithic product.

What to look for

Make sure the software supports:

• precise timestamp alignment
• offline replay from recorded data
• sensor calibration and coordinate frames
• track-level fusion or measurement-level fusion
• extensibility for custom radar and telemetry formats
• deterministic processing for reproducibility

If you want, I can narrow this down to:

1. open-source only,
2. production/commercial, or
3. best for a specific platform like Linux edge, Jetson, or Windows.