WIP Release detections

[leshy]

No description provided.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[greptile-apps]

Greptile Summary

This PR introduces performance optimizations for detection and navigation systems, including a migration to tensor-based point cloud processing and numba-accelerated occupancy grid generation.

Key Changes:

• Migrated PointCloud2 from legacy Open3D geometry to tensor-based representation for improved performance
• Added numba JIT-compiled kernels for occupancy grid generation (_height_map_kernel, _simple_occupancy_kernel)
• Extracted controller logic from LocalPlanner into dedicated PController and PdController classes
• Renamed image input to color_image in detection modules for clarity
• Upgraded langchain packages to v1.x and added numba/llvmlite as direct dependencies
• Added e2e test infrastructure with LCM spy fixtures

Issues Found:

• SimpleOccupancyConfig includes can_pass_under, can_climb, ignore_noise, and smoothing fields that are never used in simple_occupancy() function
• closing_iterations and closing_connectivity fields remain in config but morphological closing logic was removed

Confidence Score: 3/5

• This PR is reasonably safe to merge with minor cleanup recommended
• Score reflects good performance improvements but includes unused config fields that create confusion. The core logic changes are sound (numba optimization, tensor-based point clouds, controller extraction), but the SimpleOccupancyConfig has fields that suggest functionality that was removed.
• Pay close attention to dimos/mapping/pointclouds/occupancy.py - unused config fields should be removed or the corresponding functionality should be implemented

Important Files Changed

Filename	Overview
dimos/mapping/pointclouds/occupancy.py	Added numba-accelerated kernels for occupancy grid generation, removed morphological closing logic, added config fields that are not used in simple_occupancy
dimos/msgs/sensor_msgs/PointCloud2.py	Migrated from legacy Open3D geometry to tensor-based representation for performance, added backward compatibility with pickling and caching
dimos/navigation/replanning_a_star/controllers.py	New file extracting PController and PdController classes for robot motion control with rotation-then-drive behavior
pyproject.toml	Upgraded langchain packages to v1.x, moved numba/llvmlite from override-dependencies to direct dependencies, changed unitree-webrtc-connect git source

Sequence Diagram

sequenceDiagram
    participant Robot as Unitree GO2
    participant WebRTC as WebRTC Connection
    participant Detection as Detection3D/ObjectDB
    participant PointCloud as PointCloud2
    participant Occupancy as Occupancy Mapper
    participant Planner as Local Planner
    participant Controller as PController/PdController

    Robot->>WebRTC: Odometry & Camera Data
    WebRTC->>Detection: color_image stream
    WebRTC->>PointCloud: raw pointcloud
    PointCloud->>PointCloud: Convert to tensor format
    PointCloud->>Occupancy: as_numpy() (fast, no conversion)
    Occupancy->>Occupancy: Upcast to float64
    Occupancy->>Occupancy: _simple_occupancy_kernel (numba)
    Occupancy->>Planner: OccupancyGrid
    Planner->>Planner: Compute path
    Planner->>Controller: advance(lookahead_point, odom)
    Controller->>Controller: angle_diff(desired, current)
    Controller->>Controller: _compute_angular_velocity
    Controller->>Robot: Twist command

Loading

[greptile-apps]

_{35 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

logic: unused config fields in SimpleOccupancyConfig

Suggested change

	can_pass_under: float = 0.6
	can_climb: float = 0.15
	ignore_noise: float = 0.05
	smoothing: float = 1.0
	closing_iterations: int = 1
	closing_connectivity: int = 2