Get the local IP address for building connection URLs.

Convert grid cell to world coordinates (cell center).

Get the estimated gyroscope biases [bias_x, bias_y] in rad/s.

Internal H.264 capture loop.

Handle incoming CAN frame and generate response

Handle a discovery frame.

Check if nvarguscamerasrc is available (Jetson-specific).

Check if nvv4l2h264enc (Jetson hardware H.264 encoder) is available.

Send a heartbeat with current state.

Get grid height in cells.

* Map a height (Y in Three.js, meters) to an RGB color. * Blue (ground) → cyan → green → yellow → red (high). * Standard approach for LiDAR visualiz

* Height-based coloring: blue (ground) -> cyan -> green -> yellow -> red (high). * Same gradient as PointCloud3D.tsx.

Huber robust loss weight function. Returns 1.0 for residuals within threshold (quadratic region), decreasing weight for larger residuals (linear regi

Generate IMU readings from physics state. In 2D simulation: - gyro_z = angular velocity (yaw rate) - accel_z = gravity (9.81 m/s²) - other axes are z

Check if rover is within world bounds.

Inflate obstacles by the inflation radius using distance transform. Uses a two-pass distance transform (O(n) where n = total cells) instead of the na

Initialize logging with stdout and rolling file output. Returns a guard that must be held for the lifetime of the program to ensure logs are properly

Run all initialization and return the context the control loop needs plus the log guard (must outlive the daemon).

* Interpolate between two poses using Hermite spline.

Catmull-Rom interpolation between p1 and p2 with control points p0, p3.

Intersect a horizontal line with the edges of a polygon, returning X coordinates of all intersection points.

Check if the rover should be driving (motors enabled).

Check if recording is globally enabled (can start sessions). Returns false if recording was disabled by config or `--no-recording`.

Scan Annex B byte stream for IDR NAL units (type 5).

Check if the watchdog has timed out.

Check if the bounding box is valid (non-zero area).

Check if bounds are valid (have been expanded at least once)

Get keyframe poses as (x, y, theta) tuples for visualization.

Get last collision result.

Load state from disk

Load the default policy (latest version or specified by name).

Load points from a PCD file.

Load MLP layers and training metrics from a policy JSON file. Args: policy_path: Path to the policy JSON (output of train_bc.py). Re

Load poses from poses.csv file. Format: timestamp_secs,x,y,z,qx,qy,qz,qw

Load session data for splatting. Returns: points: (N, 3) array of LiDAR points in world frame poses: List of camera pose

Log operating mode change.

Log motor telemetry (currents and temperatures).

Log wheel odometry deltas.

Log robot pose (position and heading).

Log power status.

Log SLAM keyframe poses as 2D points.

Log SLAM status metrics.

Log SLAM-corrected robot trajectory.

Set log standard deviation for stochastic inference.

Log tool state.

Log commanded and actual velocity.

Find the angle of the longest edge of the polygon (heuristic for optimal sweep).

Get loop closure count.

Main worker loop.

Create a heartbeat frame with optional fault flag

Mark a cell as occupied at the given world position, updating the height envelope. Unlike raytrace, this only sets the endpoint cell to occupied with

Get mast hole position (for placing sensor mast)

Merge a session into an existing map

Advance to next bottom page (instant)

Normalize angle to [-pi, pi].

Normalize RGB u8 to CHW f32 with ImageNet mean/std.

Get current timestamp in milliseconds (truncated to u32).