habitat nav integration (habinav) by leshy · Pull Request #4011 · dimensionalOS/dimos

leshy · 2026-09-08T21:08:57Z

dimos run habitat-voxel
dimos run habitat-raycaster
dimos run habitat-nav

…ints Runs photorealistic scanned indoor scenes (HM3D) as a dimos robot: Twist in, RGB-D + pose + a registered scan out. Enough to exercise mapping, planning and following against real building geometry without a robot. habitat-sim publishes python 3.9 conda builds only, so it cannot share the dimos interpreter. It runs as a NativeModule subprocess and speaks dimos over zenoh via dimos_lcm, the standalone message package -- no dimos import on that side, which a test enforces. The LCM runtime is deliberately not installed: we publish over zenoh, and building it needs a CMake/JDK toolchain the shell does not have. nix/install.sh is the build_command. It creates the conda env, installs the deps, fetches the annotated HM3D example scene and writes the wrapper whose existence NativeModule uses as the build sentinel, so a fresh checkout builds on first run. The wrapper is needed because CLI args are appended after the executable, leaving no place for a script path. Blueprints layer so a failure can be bisected by dropping a level: habitat-teleop sim + streams, drive it with the viewer keys habitat-raycaster + RayTracingVoxelMap on a sensor-frame scan habitat-nav + MLSPlannerNative + BasicPathFollower, goal by clicking habitat-voxel VoxelGridMapper alternative on a pre-registered scan The two mappers want the scan in different frames, so habitat-voxel is an alternative rather than a layer: VoxelGridMapper assumes world-frame clouds, while RayTracingVoxelMap raytraces from the sensor origin and registers the cloud itself through tf. No MovementManager. Its arbitration is real-robot safety behaviour that fights the viewer: releasing a movement key publishes a zero twist, which latches teleop, cancels the active goal with a NaN and blocks nav_cmd_vel for a cooldown. Viewer teleop and the follower both publish onto cmd_vel instead, and the viewer's clicked point is already a PointStamped, so it needs no translation to reach the planner's goal. frames.py owns the habitat(y-up) to dimos(z-up) conversion, tested against basis vectors and a quaternion recorded from the live simulator. The camera hangs off a camera_optical frame: a pinhole points down its own +z, which in the body frame is straight up.

The aihabitat conda channel has no aarch64 habitat-sim, so advertising aarch64-linux in the flake only turned a clear "unsupported" into a confusing conda solve failure on Jetson. The build path was verified cold: env, data and wrapper moved aside, the exact build_command re-run from the four committed files, and the result driven through NativeModule to ready. No compiler step in the log.

…it from the index

codecov · 2026-09-08T21:19:21Z

❌ 7 Tests Failed:

Tests completed	Failed	Passed	Skipped
5392	7	5385	102

View the top 1 failed test(s) by shortest run time

dimos.utils.test_shm::test_create_or_attach_splits_owner_and_reader

Stack Traces | 0.508s run time

name = 'dimos_test_691cfbb67917'
slow_ftruncate = <function slow_ftruncate.<locals>.delayed at 0x7fe432caa200>

    def test_create_or_attach_splits_owner_and_reader(name, slow_ftruncate):
        """Concurrent symmetric callers: exactly one owns, the other waits and attaches."""
        out: list[tuple[bool, int]] = []
        lock = threading.Lock()
        barrier = threading.Barrier(2)
    
        def racer() -> None:
            barrier.wait()
            shm, owner = create_or_attach_shm(name, SIZE, timeout=5.0)
            with lock:
                out.append((owner, shm.size))
            shm.close()
    
        threads = [threading.Thread(target=racer, daemon=True) for _ in range(2)]
        for t in threads:
            t.start()
        for t in threads:
            t.join(timeout=8)
    
        assert len(out) == 2, f"a racer did not finish: {out}"
        assert sorted(owner for owner, _ in out) == [False, True]
>       assert {size for _, size in out} == {SIZE}
E       assert {0, 65536} == {65536}
E         
E         Extra items in the left set:
E         #x1B[0m#x1B[94m0#x1B[39;49;00m#x1B[90m#x1B[39;49;00m
E         
E         Full diff:
E         #x1B[0m#x1B[90m #x1B[39;49;00m {#x1B[90m#x1B[39;49;00m
E         #x1B[92m+     0,#x1B[39;49;00m#x1B[90m#x1B[39;49;00m
E         #x1B[90m #x1B[39;49;00m     65536,#x1B[90m#x1B[39;49;00m
E         #x1B[90m #x1B[39;49;00m }#x1B[90m#x1B[39;49;00m

barrier    = <threading.Barrier at 0x7fe3d5ed61b0: waiters=0/2>
lock       = <unlocked _thread.lock object at 0x7fe432bf3b00>
name       = 'dimos_test_691cfbb67917'
out        = [(True, 65536), (False, 0)]
racer      = <function test_create_or_attach_splits_owner_and_reader.<locals>.racer at 0x7fe432ca9760>
slow_ftruncate = <function slow_ftruncate.<locals>.delayed at 0x7fe432caa200>
t          = <Thread(Thread-7094 (racer), stopped daemon 140616151324352)>
threads    = [<Thread(Thread-7093 (racer), stopped daemon 140614410696384)>, <Thread(Thread-7094 (racer), stopped daemon 140616151324352)>]

dimos/utils/test_shm.py:149: AssertionError

View the full list of 7 ❄️ flaky test(s)

dimos.experimental.memory.test_rust_recorder_e2e::test_cli_recording_uses_existing_binary_for_both_formats[mcap]