feat(mapping): add a voxel clear mask so a sensor can erase its own ghost - #3712
Merged
Conversation
mustafab0
requested review from
aclauer,
paul-nechifor and
spomichter
as code owners
August 27, 2026 01:00
mustafab0
force-pushed
the
mustafa/feat/voxel-clear-mask
branch
from
August 27, 2026 01:02
d713c5a to
3502047
Compare
Contributor
Greptile SummaryThis change adds a point-cloud clear mask for removing mapped voxels and updates the mapper’s persistent and live-overlay data. Rust tests reproduced that clearing a healthy voxel leaves it eligible for point emission and leaves neighboring support counts inflated, so the clear operation does not fully remove the voxel’s map presence. Confidence Score: 4/5Not safe to merge until voxel clearing synchronizes the healthy-chunk index and neighboring support counters. Focused Rust execution exercised the healthy-voxel clearing path twice and consistently showed that a deleted voxel remains emitted and that a neighbor retains excess support. Files Needing Attention: dimos/mapping/ray_tracing/rust/src/voxel_ray_tracer.rs
 T-Rex LogsWhat T-Rex did
|
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Codecov Report✅ All modified and coverable lines are covered by tests. @@ Coverage Diff @@
## main #3712 +/- ##
==========================================
- Coverage 78.01% 78.00% -0.01%
==========================================
Files 1301 1301
Lines 125328 125329 +1
Branches 10952 10952
==========================================
- Hits 97769 97765 -4
- Misses 24375 24381 +6
+ Partials 3184 3183 -1
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 3 files with indirect coverage changes 🚀 New features to boost your workflow:
|
…host Ray tracing only clears what a ray passes through. A sensor that occludes itself - a wrist camera staring past its own arm - can never clear the volume its arm hides: no ray reaches through the arm, so record_miss never fires there. The arm's own returns land in the map and stay, and over a work session the robot slowly walls itself in. A publisher that knows those points are free now says so on a voxel_clear_mask port. The mask is world-frame metric points, stamp-ordered and gated against world_frame, quantized through the same path returns take. VoxelMap::clear_voxels deletes by key, not with a bare remove: it also pulls the key out of the healthy-chunk index emit_points scans and decrements every existing 26-neighbour's support counter, exactly as record_miss does on its removal path. Skipping either leaves an index pointing at a voxel that is gone and support counts that never come back down. An unhealthy voxel was never in its neighbours' counts, so removing it leaves them alone. The fine-cell bitmask rides inside the removed voxel, so the fine layer needs no separate cleanup. Mapper::clear_metric drops this frame's live hits alongside the voxels, so a cleared voxel cannot come back out of the live overlay.
mustafab0
force-pushed
the
mustafa/feat/voxel-clear-mask
branch
from
August 27, 2026 01:33
3502047 to
787ae2c
Compare
1 task
aclauer
reviewed
Aug 27, 2026
aclauer
left a comment
aclauer
left a comment
Collaborator
There was a problem hiding this comment.
Does setting the shadow_depth config not fix this? It defaults to 0.1
mustafab0
added a commit
that referenced
this pull request
Aug 27, 2026
Manipulation becomes the first non-navigation consumer of the ray-tracing voxel map. A wrist camera builds an occupancy map of the workspace, and the planner treats those voxels as one octree obstacle. Three producers, all model-driven so they work for any URDF: - RobotTfPublisher resolves a complete TF tree from measured joint state. A manipulator's sensor pose comes from forward kinematics, not SLAM, and main had no generic equivalent - only a G1-specific publisher. An incomplete joint state publishes nothing rather than leaving links at a stale pose. - PointCloudSelfFilter cuts the robot out of its own camera's cloud, and emits the map cells the robot occupies as a clear mask. It emits the previous frame's cells too: a link that moved leaves a ghost behind it that ray tracing can never clear, because the link occluded that volume while it was there. The mask floors world coordinates by voxel_size exactly as the mapper does, or it would name cells the map does not hold. - GlobalMapObstacleBridge reconciles each complete map snapshot into the planning world under one stable obstacle ID, so a republished map replaces the obstacle instead of piling up new ones. ObstacleType.OCTREE carries occupied cell centers plus a resolution into roboplan's addOcTreeGeometry. Coal reads a cell as center, edge length, cost and occupancy threshold; every cell here is occupied, so cost is 1.0 against the 0.5 default. Verified against roboplan 0.5.1 and the pinned 0.6.1: adding cells flips a collision-free configuration to colliding and blocks a path through them, while a clear path stays clear. Obstacle.points is a tuple of tuples, not an ndarray. Obstacles are deepcopied into world snapshots, compared, and pickled across worker RPC, and an ndarray field breaks all three. The shape-string obstacle RPCs cannot carry a map, so ManipulationModule gains set_voxel_obstacle alongside them. Additive: no existing caller changes. An empty point list removes the obstacle, which is how a mapper says the space it owns is clear. Validation caps an octree at 200k points, since every point crosses worker RPC as a pickled tuple and an unbounded map would stall the pipe. Depends on the voxel_clear_mask port (#3712) for the mask to land anywhere, and on the TFLookup forward_tolerance keyword (#3711) for the self filter to wait on a wrist transform still in flight.
github-actions
Bot
added
the
ready-to-merge
aclauer
approved these changes
Aug 27, 2026
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Aug 27, 2026
mustafab0
added a commit
that referenced
this pull request
Aug 27, 2026
Manipulation becomes the first non-navigation consumer of the ray-tracing voxel map. A wrist camera builds an occupancy map of the workspace, and the planner treats those voxels as one octree obstacle. Three producers, all model-driven so they work for any URDF: - RobotTfPublisher resolves a complete TF tree from measured joint state. A manipulator's sensor pose comes from forward kinematics, not SLAM, and main had no generic equivalent - only a G1-specific publisher. An incomplete joint state publishes nothing rather than leaving links at a stale pose. - PointCloudSelfFilter cuts the robot out of its own camera's cloud, and emits the map cells the robot occupies as a clear mask. It emits the previous frame's cells too: a link that moved leaves a ghost behind it that ray tracing can never clear, because the link occluded that volume while it was there. The mask floors world coordinates by voxel_size exactly as the mapper does, or it would name cells the map does not hold. - GlobalMapObstacleBridge reconciles each complete map snapshot into the planning world under one stable obstacle ID, so a republished map replaces the obstacle instead of piling up new ones. ObstacleType.OCTREE carries occupied cell centers plus a resolution into roboplan's addOcTreeGeometry. Coal reads a cell as center, edge length, cost and occupancy threshold; every cell here is occupied, so cost is 1.0 against the 0.5 default. Verified against roboplan 0.5.1 and the pinned 0.6.1: adding cells flips a collision-free configuration to colliding and blocks a path through them, while a clear path stays clear. Obstacle.points is a tuple of tuples, not an ndarray. Obstacles are deepcopied into world snapshots, compared, and pickled across worker RPC, and an ndarray field breaks all three. The shape-string obstacle RPCs cannot carry a map, so ManipulationModule gains set_voxel_obstacle alongside them. Additive: no existing caller changes. An empty point list removes the obstacle, which is how a mapper says the space it owns is clear. Validation caps an octree at 200k points, since every point crosses worker RPC as a pickled tuple and an unbounded map would stall the pipe. Depends on the voxel_clear_mask port (#3712) for the mask to land anywhere, and on the TFLookup forward_tolerance keyword (#3711) for the self filter to wait on a wrist transform still in flight.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The physics
Ray tracing clears a voxel by passing a ray through it to a farther return. That works when the sensor sees past the thing it is mapping.
A wrist camera does not. Its own arm sits between the lens and the space behind it, so no ray ever reaches through the arm and
record_missnever fires in that volume. Meanwhile the arm's own returns land in the map as ordinary hits. The result: the arm deposits voxels of itself, and there is no mechanism that can ever erase them. Over a work session the robot slowly walls itself in with a ghost of where it has been, and a planner readingglobal_maprefuses to go there.Ray-trace clearing cannot fix this — the geometry forbids it. The information has to come from somewhere else: a publisher that knows, from URDF plus TF, that a given volume is the robot and therefore free.
What this adds
A
voxel_clear_mask: Input<PointCloud2>port. The cloud is world-frame metric points meaning "these are definitely free"; the node deletes the voxels covering them.world_frame. The points are metric positions, so the mask must already be in the world frameTests
cargo test --workspace --all-features --locked: 60 passed.cargo fmt --checkandcargo clippy -D warningsclean..Files: 6.