A clear path from setup to review.

Use these complete packages for the corrected local viewer and optional trainer. Verify the SHA-256 companion and extract the whole folder. These files are hosted here and in the verified public GitHub prereleases: viewer v0.9.0-preview.3 and trainer v0.1.0-preview.2.

• Radar Studio v0.9.0-preview.3 viewer ZIP (396733 bytes) · SHA-256 file · exact viewer source commit. Extract the complete radar-studio/ folder.
• Original radar host training tools v0.1.0-preview.2 ZIP (22097 bytes) · SHA-256 file · exact recorder source commit. Extract original_training/; Python 3.10+ standard library only.

Neither ZIP is firmware to flash. Python is optional host analysis software and is not needed to open this viewer. No participant recordings or trained radar weights ship. Full checksums, setup and historical archives. The separate original sensing firmware 0.3.1-rc1 and older GitHub releases retain their existing bytes and gates. Initial no-RF receipt review still precedes separately authorized manual sensing.

This site contains the browser application from Radar Studio v0.9.0-preview.5. It is a hardware prototype with experimental sensing and learning workflows. Hosting does not add firmware support, hardware validation or recognition accuracy.

1. Open the viewer and select Original radar v2 · JSON Lines · separate control + data for the original firmware workflow. Timed tests support firmware 0.3.1-rc1 / abb27b48c1bdb956c0de704b. Continuous mode requires 0.3.2-rc1 / 87589b2c32638ae119334c72. Old one-port firmware cannot unlock this workflow.
2. Choose Enhanced control at 115200 baud and the separate Standard data at 921600 baud on the same board, both 8N1. The data lane is read-only to the host. Query control status first; require coherent LF-delimited status/ACK, passing version-2 self-test, exact identity and fault-free idle. Connecting never starts RF.
3. Enter your measured bracket-top height, tilt and optional signed vertical antenna offset. Height defaults to 1219 mm and tilt to 0°; unknown antenna offset remains blank and measurement confirmation starts unchecked. These defaults are not verified measurements: check and correct them for your actual setup. Height means floor to the flat top of the bracket, excluding screws. Confirm the measurement checkbox. Unknown offset stays unknown; declarations are not physical calibration.
4. For the initial timed check, choose an integer 1–5 second duration, then Configure only. Click Verify port pairing: a fresh control-issued nonce must appear on the separate data lane before pair confirmation and fresh matching control status. This correlates lanes; it is not authentication. Configuration and pairing start no RF.
5. Finish the initial no-RF checkpoint here. Save the capture JSON and both separately labelled raw UART streams. Stop and review the status, self-test, configuration and nonce receipts before considering sensing. A passed browser gate does not authorize the next hardware step.
6. Only after that receipt review, separate authorization, and all identity, self-test, fault, mounting, current-connection/configuration and nonce gates pass can a separate manual Start run the bounded test. Keep control connected for Stop/status. Failed pairing, malformed/wrong-lane data or reconnects invalidate readiness. A software review is not hardware, calibration or sensing-accuracy validation. Complete original dual-UART guide.
7. Use a Web Serial-capable desktop browser and open the site directly over HTTPS. Chrome/Edge desktop are the intended browsers; mobile or browsers without Web Serial cannot use live USB serial here. Managed-device restrictions may block access. Official browser guidance. Offline tools remain available for compatible files.
8. The original candidate 3D display shows at most three firmware peaks per frame. Temporary candidate IDs and measured trails are experimental reflection associations, not a people count, stable human identity, posture classifier or full people point cloud.
9. Existing TI workflows remain separate and unchanged. Select the exact TI catalogue profile only for matching TI firmware. Their synthetic demos are not validation of the original pipeline. TI firmware/configuration comes from Radar Toolbox, with flashing through UniFlash outside this site. Follow board documentation.

Open the site directly, rather than in an embedded preview, when selecting serial devices. Inspect the app's compatibility notice if access is unavailable. Keep another app from claiming the same serial ports.

Firmware 0.3.2-rc1 supports Run until stopped, with a five-second exact-session keepalive lease. Reconfigure and verify pairing for the chosen mode, then use the separate manual Start. Use local folder recording to preserve a full run; the diagnostic download is only a bounded window. Recording, camera review and Python training instructions. The browser trainer accepts earlier TI-format data and cannot import the new recording folder or camera review JSON.

Choose Enhanced control → choose Standard data → Query control status → check measured height/tilt and confirm → Configure only → Verify port pairing → save capture and both raw streams → stop for initial no-RF receipt review. Only after separate authorization: repeat the fresh gates → separate manual Start for 5 seconds or less → confirm final idle → save both raw streams and capture. Configuration and pairing never start RF. Start remains gated by the exact reviewed build, version-2 self-test, fault-free fresh status, current configuration/connection and matching data nonce.

If a command times out, save the capture, disconnect and reconnect control, query fresh status, then repeat Configure and Verify port pairing. Resetting only the board does not clear browser desynchronization. Do not bypass a timeout or failed gate.

The receiver polarity correction is documented hardware-convention compensation, not measured per-device phase calibration or verified physical sensing. Display range defaults to 20 m with 2/5/10/20 m choices and zoom controls. This changes only the viewport, not the RF profile. The unchanged search bins span approximately 0.8–5.9 m; physical human-detection range remains unvalidated. A 20 m view is not a 20 m sensing claim.

Prepare a named session and a walking, standing, sitting or transition practice prompt before frames arrive. Prompts describe intent, not observed truth. Guide controls never configure the device or start RF. Finish the initial no-RF receipt checkpoint before separately authorized sensing. Every later take still follows the existing measured-mount, status, Configure, nonce pairing and separate manual 1–5 second Start sequence. Confirm idle, save the capture and both raw streams, then use a fresh tab after disconnecting for a new named session.

Open local session replay and review to load an original dual-UART capture, scrub range/velocity observations, inspect gaps/faults/truncation and create reviewed activity or transition intervals. The importer corroborates admitted records against preserved raw bytes; imported provenance remains a claim, not authenticated device proof. Original files and live annotations are never rewritten. Unknown, unreviewed and excluded labels remain untrainable. Save the derived dataset to retain review work, and reload it only with the matching original capture.

Derived exports include selected observations, review labels, session metadata and hashes; raw UART bytes and video are excluded. Aliases and notes can still be private. There are no automatic uploads. Full workflow and dataset contract.

The separate original-radar Python trainer has a v0.1.0-preview.3 camera-import host-tool ZIP in the download section above, with exact source and checksum links. The earlier GitHub download remains historical; this is a host tool, not firmware. Python is not required for this viewer. Reviewed labels, enough clean temporal context and independent session/participant groups are required for evaluation. No real-world activity accuracy is established. Optional camera selection, local recording, pose-assisted human review and matched-event clock fitting are now available. Camera workflow and limits. C920e is the test device; other cameras are selectable but unverified.

All twelve catalogue profiles remain available. “Partial” means selected protocol fields and views, never the complete TI demo application. The Original radar path has a documented first brief hardware recording; the TI catalogue paths and recognition accuracy remain unverified in this hosted distribution.

• Partial protocol: Out of Box; 3D People Tracking; Low-Power 3D People Tracking; Overhead 3D People Tracking; Vital Signs with People Tracking; Small Obstacle Detection. Known points, tracks and relevant vitals are retained; not all TLVs, heatmaps, occupancy decisions or scene transforms are implemented.
• Partial protocol: Area Scanner and Automated Doors and Gates. Dedicated 44-byte framing, dynamic/static points and short tracks. No warning zones, actuator control or door-state prediction.
• Partial protocol: Child Presence Detection with Classification. Dedicated 48-byte checksum framing and raw compressed point clouds. Classification and vehicle/cuboid transformations require TI's external MATLAB tooling. This site makes no child-presence or safety decision.
• Experimental: Gesture Recognition with ML. Raw feature/probability telemetry, not TI's postprocessed gesture decisions.
• Experimental: True Ground Speed. Surface-velocity telemetry, validity, freshness and history.
• Experimental: High Accuracy Level Sensing. Dedicated 36-byte framing, Q20 range/history and selected FFT-input magnitude. No tank volume or calibrated range spectrum.

See the full catalogue and exact BIN/CFG metadata or machine-readable catalogue. Unknown or mismatched formats must not be treated as supported. Local artifact inspection checks metadata and hashes, never flashes or configures a board.

• TI two-port serial workflow, acknowledgement-checked configuration, local .bin replay and profile-specific synthetic demonstrations
• 3D point/track view, orbit/top/front camera controls, movement trails, vitals retention and protocol-specific telemetry panels
• Motion Signature Lab: measured coordinate calibration, one-target gating, observed activity and transition annotations, local training, held-out evaluation, dataset import/export and complete received-UART capture
• Offline ML: local recordings/features, independent capture splits, activity/event models, saved models, held-out metrics, proposals and separate human-review overlays. Workflow details.
• Original radar v2: separate control/data JSON Lines transport; exact-build and current-session gates; measured mounting; nonce pairing; explicit status/configure/start/stop; and experimental sensor-coordinate 3D candidate association. Save separate exact-byte control and data JSONL files, or capture JSON with separately labelled raw streams, bounded admitted frames, local derived tracking, drop counts and human annotations. Raw lanes are never rewritten or interleaved. No people count, posture model, stable identity, calibrated confidence or world-coordinate accuracy is claimed. Timed tests remain available; continuous mode requires the new firmware and a healthy heartbeat. Diagnostic retention is bounded; full local recording is separate.

The hosted viewer needs a compatible Web Serial browser and board access, not a local TI SDK, CCS or Python installation. Use the official board page and hardware guides, TI UniFlash for separate flashing, and official Silicon Labs CP210x drivers if required by your operating system. TI Radar Toolbox is for the separate TI demo workflow; the original firmware is a separate reviewed project build. See download details and exact package paths. No vendor binaries or drivers are redistributed here.

Firmware compilation/flashing, TI SDK/CCS/UniFlash, vendor GUI parity, CPD classification, raw ADC acquisition and automatic camera synchronization are not browser features. The optional Node.js offline CLI remains available in the separate source project; it is not executed by this website. A third-party pose landmark model is bundled for camera review; no pretrained radar activity model, cloud training or private recording is supplied.

Motion and offline classifiers are experimental local baselines. Human annotations are not automatic ground truth. Unknown calibration, uncertain windows and invalid data retain their original rejection safeguards. Vitals and all other estimates are not medical measurements or safety decisions.

Original firmware mounting uses your measured bracket-top height, tilt and optional signed antenna offset. Unknown values remain unknown; declarations do not establish calibrated world coordinates.

Opening the viewer downloads its application files; account and contribution preview pages are served separately. Selected files, serial bytes, annotations and learned parameters stay in your browser until you explicitly download them; normal viewer use sends no recordings. Separate contribution routes exist, but recording intake is compiled off. The hosting provider may receive ordinary access requests. This is not a claim about browser extensions or device security.

Reloading or closing the page loses unsaved work. Export before leaving. The viewer’s persistent safety value is the Original radar unconfirmed-stop boolean; it contains no recordings. Separate account pages use their own sign-in/session handling. Closing a tab or disconnecting cannot guarantee RF stopped. Use Stop and confirm idle. Save does not stop RF. Obtain participant consent and review exported records before sharing. Recording intake and public contributions are not enabled. Optional contribution preview and account/privacy-request pages remain separate. Published terms and privacy notices do not enable intake. No cloud inference or model-training job is provided.

Free access under the Radar Studio Chromebook Source-Available No-Resale License. Resale and paid software access are restricted; this is not an OSI-approved open-source license. Version coverage and attribution / third-party exclusions are preserved. TI material remains under its own terms and is not bundled or relicensed.

This hosted distribution adds this guide, a redesigned workspace, hosting/privacy notices and public-facing documentation edits. The hosted workspace retains the v0.9.0-preview.5 transport and recording behavior and adds interface and keyboard-accessibility improvements. The downloadable preview.5 ZIP remains its original versioned snapshot. Catalogue metadata and earlier protocol support are retained. The hosted adaptation does not weaken or bypass any firmware gate. Personal setup history and research planning are excluded.