feat(perception): StereoPointCloud module + FlowBase stereo-nav blueprint #2768
+696
−0
There are no files selected for viewing
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
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
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,241 @@ | ||
| # Copyright 2025-2026 Dimensional Inc. | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| """Stereo depth → per-frame cloud + persistent global map for CostMapper.""" | ||
|
|
||
| from __future__ import annotations | ||
|
|
||
| import threading | ||
| from typing import Any | ||
|
|
||
| import numpy as np | ||
| from reactivex.disposable import Disposable | ||
| from scipy.ndimage import sobel | ||
|
|
||
| from dimos.core.core import rpc | ||
| from dimos.core.module import Module, ModuleConfig | ||
| from dimos.core.stream import In, Out | ||
| from dimos.msgs.sensor_msgs.CameraInfo import CameraInfo | ||
| from dimos.msgs.sensor_msgs.Image import Image | ||
| from dimos.msgs.sensor_msgs.PointCloud2 import PointCloud2 | ||
| from dimos.utils.logging_config import setup_logger | ||
|
|
||
| logger = setup_logger() | ||
|
|
||
| # ── Constants ───────────────────────────────────────────────────────────────── | ||
|
|
||
| _VOFF = np.int64(100_000) | ||
| _VMASK = np.int64(0x3FFFF) | ||
|
|
||
|
|
||
| # ── Helpers ─────────────────────────────────────────────────────────────────── | ||
|
|
||
| def _pack(vkeys: np.ndarray) -> np.ndarray: | ||
| v = (vkeys.astype(np.int64) + _VOFF) & _VMASK | ||
| return (v[:, 0] << 36) | (v[:, 1] << 18) | v[:, 2] | ||
|
|
||
|
|
||
| def _gradient_mask(depth: np.ndarray, threshold: float) -> np.ndarray: | ||
| """True where depth is valid and Sobel gradient magnitude is below threshold (m/px).""" | ||
| valid = np.isfinite(depth) | ||
| depth_f = np.where(valid, depth, 0.0).astype(np.float64) | ||
| grad = np.hypot(sobel(depth_f, axis=1), sobel(depth_f, axis=0)) | ||
| return valid & (grad < threshold) | ||
|
|
||
|
|
||
| def _raycast_free_keys( | ||
| origin: np.ndarray, | ||
| surface_pts: np.ndarray, | ||
| vox: float, | ||
| n_rays: int = 200, | ||
| max_steps: int = 40, | ||
| ) -> np.ndarray: | ||
| """World-frame voxel keys on rays from origin to surface_pts. Keys match floor(pts/vox) world voxelisation.""" | ||
| if len(surface_pts) == 0: | ||
| return np.array([], dtype=np.int64) | ||
|
|
||
| idx = np.random.choice(len(surface_pts), min(n_rays, len(surface_pts)), replace=False) | ||
| vecs = (surface_pts[idx] - origin).astype(np.float32) | ||
| dist = np.linalg.norm(vecs, axis=1, keepdims=True) | ||
| dist = np.where(dist < 1e-6, 1e-6, dist) | ||
| dirs = vecs / dist | ||
|
|
||
| steps = np.linspace(0.1, 0.9, max_steps, dtype=np.float32) | ||
| pts = origin + dirs[:, None, :] * (dist[:, None, :] * steps[None, :, None]) | ||
| pts = pts.reshape(-1, 3) | ||
| return np.unique(_pack(np.floor(pts / vox).astype(np.int32))) | ||
|
|
||
|
|
||
| # ── Module ──────────────────────────────────────────────────────────────────── | ||
|
|
||
| class Config(ModuleConfig): | ||
| min_depth: float = 0.3 | ||
| max_depth: float = 8.0 | ||
| gradient_threshold: float = 0.30 # m/px; above → depth edge artifact → reject | ||
| vox_size: float = 0.020 # per-frame dedup voxel (2 cm) | ||
| global_vox_size: float = 0.040 # accumulation voxel (4 cm, absorbs pose jitter) | ||
| floor_margin: float = 0.04 # metres above detected floor to keep points | ||
| max_global_pts: int = 200_000 | ||
| publish_every: int = 3 # emit global_map every N frames | ||
| world_frame: str = "world" | ||
| camera_frame: str = "camera_color_optical_frame" | ||
| base_frame: str = "base_link" | ||
| tf_timeout: float = 0.2 | ||
|
|
||
|
|
||
| class StereoPointCloud(Module): | ||
| """Gradient filter → backproject → floor removal → voxel dedup → ray-cast ghost clearing → global_map.""" | ||
|
|
||
| config: Config | ||
|
|
||
| depth_image: In[Image] | ||
| depth_camera_info: In[CameraInfo] | ||
|
|
||
| frame_cloud: Out[PointCloud2] | ||
| global_map: Out[PointCloud2] | ||
|
|
||
| def __init__(self, **kwargs: Any) -> None: | ||
| super().__init__(**kwargs) | ||
| self._lock = threading.Lock() | ||
| self._latest_info: CameraInfo | None = None | ||
| self._last_tf = None | ||
| self._floor_buf: list[float] = [] | ||
| self._acc_pts: np.ndarray = np.empty((0, 3), dtype=np.float32) | ||
| self._frame = 0 | ||
|
|
||
| @rpc | ||
| def start(self) -> None: | ||
| super().start() | ||
| self.register_disposable(Disposable(self.depth_camera_info.subscribe(self._on_info))) | ||
| self.register_disposable(Disposable(self.depth_image.subscribe(self._on_depth))) | ||
|
|
||
| @rpc | ||
| def stop(self) -> None: | ||
| super().stop() | ||
|
|
||
| def _on_info(self, info: CameraInfo) -> None: | ||
| with self._lock: | ||
| self._latest_info = info | ||
|
|
||
| def _on_depth(self, img: Image) -> None: | ||
| with self._lock: | ||
| info = self._latest_info | ||
|
|
||
| depth = img.data | ||
| if hasattr(depth, "get"): | ||
| depth = depth.get() | ||
| if depth.ndim == 3: | ||
| depth = depth[:, :, 0] | ||
| depth = depth.astype(np.float32) | ||
| if np.nanmedian(depth[depth > 0]) > 100: | ||
| depth /= 1000.0 | ||
|
|
||
| H, W = depth.shape | ||
|
|
||
| if info is not None: | ||
| K = info.get_K_matrix() | ||
| fx, fy, cx, cy = float(K[0, 0]), float(K[1, 1]), float(K[0, 2]), float(K[1, 2]) | ||
| else: | ||
| fx = fy = float(max(H, W)) / 2.0 | ||
| cx, cy = W / 2.0, H / 2.0 | ||
|
|
||
| mask = ( | ||
| _gradient_mask(depth, self.config.gradient_threshold) | ||
| & (depth > self.config.min_depth) | ||
| & (depth < self.config.max_depth) | ||
| ) | ||
| if not mask.any(): | ||
| return | ||
|
|
||
| uu, vv = np.meshgrid(np.arange(W, dtype=np.float32), np.arange(H, dtype=np.float32)) | ||
| dd = depth[mask] | ||
| xyz_opt = np.column_stack([ | ||
| (uu[mask] - cx) * dd / fx, | ||
| (vv[mask] - cy) * dd / fy, | ||
| dd, | ||
| ]).astype(np.float32) | ||
|
|
||
| tf = self.tf.get( | ||
| self.config.world_frame, self.config.camera_frame, img.ts, self.config.tf_timeout | ||
| ) | ||
| if tf is not None: | ||
| self._last_tf = tf | ||
| if self._last_tf is not None: | ||
| R = self._last_tf.rotation.to_rotation_matrix().astype(np.float32) | ||
| t = np.array( | ||
| [self._last_tf.translation.x, | ||
| self._last_tf.translation.y, | ||
| self._last_tf.translation.z], | ||
| dtype=np.float32, | ||
| ) | ||
| else: | ||
| R, t = np.eye(3, dtype=np.float32), np.zeros(3, dtype=np.float32) | ||
|
|
||
| xyz_world = (xyz_opt @ R.T + t).astype(np.float32) | ||
|
|
||
| # TF-primary floor detection; falls back to rolling low-Z percentile | ||
| base_tf = self.tf.get( | ||
| self.config.world_frame, self.config.base_frame, img.ts, self.config.tf_timeout | ||
| ) | ||
| if base_tf is not None: | ||
|
Contributor
There was a problem hiding this comment.
|
||
| floor_z = float(base_tf.translation.z) | ||
| else: | ||
| bot = xyz_world[xyz_world[:, 2] < np.percentile(xyz_world[:, 2], 10)] | ||
| if len(bot): | ||
| self._floor_buf.append(float(np.median(bot[:, 2]))) | ||
| if len(self._floor_buf) > 30: | ||
| self._floor_buf.pop(0) | ||
| floor_z = float(np.percentile(self._floor_buf, 20)) if len(self._floor_buf) >= 5 else None | ||
|
|
||
| vk = np.floor(xyz_world / self.config.vox_size).astype(np.int32) | ||
| _, ui = np.unique(_pack(vk), return_index=True) | ||
| xyz_vox = xyz_world[ui] | ||
| if floor_z is not None: | ||
| xyz_vox = xyz_vox[xyz_vox[:, 2] > floor_z + self.config.floor_margin] | ||
| if len(xyz_vox) == 0: | ||
| return | ||
|
|
||
| self.frame_cloud.publish( | ||
| PointCloud2.from_numpy(xyz_vox, frame_id=self.config.world_frame, timestamp=img.ts) | ||
| ) | ||
|
|
||
| pts_snap = None | ||
|
|
||
| with self._lock: | ||
| if len(self._acc_pts) > 0: | ||
| free = _raycast_free_keys(t, xyz_vox, self.config.global_vox_size) | ||
| if len(free): | ||
| keys_acc = _pack(np.floor(self._acc_pts / self.config.global_vox_size).astype(np.int32)) | ||
| self._acc_pts = self._acc_pts[~np.isin(keys_acc, free)] | ||
|
|
||
| self._acc_pts = ( | ||
| np.vstack([self._acc_pts, xyz_vox]) | ||
| if len(self._acc_pts) else xyz_vox.copy() | ||
| ) | ||
| vk_g = np.floor(self._acc_pts / self.config.global_vox_size).astype(np.int32) | ||
| _, ui_g = np.unique(_pack(vk_g), return_index=True) | ||
| self._acc_pts = self._acc_pts[ui_g] | ||
|
|
||
| if len(self._acc_pts) > self.config.max_global_pts: | ||
| keep = np.random.choice(len(self._acc_pts), self.config.max_global_pts, replace=False) | ||
| self._acc_pts = self._acc_pts[keep] | ||
|
Contributor
There was a problem hiding this comment.
When the accumulated map exceeds |
||
|
|
||
| self._frame += 1 | ||
| if self._frame % self.config.publish_every == 0 and len(self._acc_pts): | ||
| pts_snap = self._acc_pts.copy() | ||
|
|
||
| if pts_snap is not None: | ||
| self.global_map.publish( | ||
| PointCloud2.from_numpy(pts_snap, frame_id=self.config.world_frame, timestamp=img.ts) | ||
| ) | ||
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
Oops, something went wrong.
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.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The documented
dimos run coordinator-flowbase-stereo-navpath resolves through the generated blueprint registry, but this change only adds the blueprint variable. Since the registry was not updated withcoordinator-flowbase-stereo-nav, the advertised command can fail before this blueprint is loaded.