Ivan/fix/go2recording

[leshy]

memory2 recorder wasn't appending poses to Observations stored.
now Recorder module uses the transform system to find global coordinates and timestamp of a message (according to its frame ID)

[greptile-apps]

Greptile Summary

This PR fixes the memory2 Recorder module, which was storing observations without their robot pose. The recorder now looks up each message's world-frame transform via self.tf at record time, attaching pose and timestamp to every Observation written to the SQLite store.

• Recorder._port_to_stream replaces the old direct stream.append(msg) with a new on_msg callback that resolves the tf transform (world → message's frame_id) and passes ts and pose to stream.append.
• RecorderConfig gains default_frame_id and tf_tolerance fields; when tf is unavailable the recording proceeds without a pose and a warning is emitted.
• Test fixtures switch from the go2_bigoffice legacy dataset to go2_short, and test_codecs.py replaces TimedSensorReplay with a direct SqliteStore read.

Confidence Score: 5/5

Safe to merge; the core recording path degrades gracefully when tf is unavailable and all observations are still written to the store.

The change is well-scoped: tf lookup failures are handled with a None-pose fallback so recording is never blocked. The only findings are a redundant field declaration in RecorderConfig and a per-message warning that could flood logs when tf is absent — neither affects correctness or data integrity.

dimos/memory2/module.py — the RecorderConfig field duplication and warning-log volume are worth a quick look before shipping to production deployments with high-frequency sensors.

Important Files Changed

Filename	Overview
dimos/memory2/module.py	Core fix: Recorder._port_to_stream now uses self.tf to attach world-frame pose to every observation; RecorderConfig gains default_frame_id, tf_tolerance, and a redundant db_path field already defined by the parent MemoryModuleConfig
dimos/memory2/codecs/test_codecs.py	Replaces TimedSensorReplay with SqliteStore to source JPEG codec test frames; get_data and SqliteStore calls are outside the skip-guard try/except block
dimos/memory2/test_e2e.py	Swaps bigoffice DB references for go2_short to match the updated LFS test fixture

Sequence Diagram

sequenceDiagram
    participant Sensor as Sensor (In port)
    participant Recorder
    participant LCMTF as self.tf (LCMTF)
    participant Stream
    participant SqliteStore

    Recorder->>Recorder: start() - register _port_to_stream per In port
    Sensor->>Recorder: on_msg(msg)
    Recorder->>Recorder: "ts = msg.ts or time.time()"
    Recorder->>Recorder: "frame_id = msg.frame_id or default_frame_id"
    Recorder->>LCMTF: "get(world, frame_id, time_point=ts, tf_tolerance)"
    alt transform found
        LCMTF-->>Recorder: Transform
        Recorder->>Recorder: "pose = transform.to_pose()"
    else transform not found
        LCMTF-->>Recorder: None
        Recorder->>Recorder: "pose = None, log warning"
    end
    Recorder->>Stream: "append(msg, ts=ts, pose=pose)"
    Stream->>SqliteStore: persist Observation(data, ts, pose)

Loading

_{Reviews (5): Last reviewed commit: "Merge branch 'dev' into ivan/fix/go2reco..." | Re-trigger Greptile}

[codecov]

Codecov Report

❌ Patch coverage is 25.00000% with 24 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
dimos/memory2/module.py	35.00%	13 Missing ⚠️
dimos/memory2/codecs/test_codecs.py	0.00%	8 Missing ⚠️
dimos/memory2/test_e2e.py	25.00%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!