Avoidance Architecture & Data Fusion

Executive Summary

ArduPilot's avoidance system is a multi-layered stack that ingests raw sensor data (Lidar, Radar, Depth Cameras), fuses it into a unified "Boundary," and then modifies the vehicle's desired velocity to prevent collisions.

The architecture separates the Sensor Driver (AP_Proximity) from the Control Logic (AC_Avoidance).

Theory & Concepts

1. The Sensor Layer (`AP_Proximity`)

This library manages the hardware drivers. It supports 360-degree Lidars (RPLidar, SF40C) and 1D Rangefinders arranged in a ring.

The Boundary: Sensors populate a "3D Boundary," which is essentially a low-resolution polar grid (sectors) storing the closest obstacle in that direction.
Fusion: If multiple sensors see an object in the same sector, the closest distance is used (Safety Conservative).

2. The Control Layer (`AC_Avoid`)

This library sits between the Pilot/Auto-Mission and the Position Controller.

Input: Desired Velocity Vector ($V_{des}$).
Logic: Checks if $V_{des}$ intersects with the Boundary.
Output: Modified Velocity Vector ($V_{safe}$).

Codebase Investigation

1. Proximity Backend: `AP_Proximity_Backend`

Located in libraries/AP_Proximity/AP_Proximity_Backend.cpp.

Drivers call update() to fetch data from UART/CAN.
They push data into the boundary using frontend.boundary.set_face_distance().

2. The Adjust Velocity Loop: `AC_Avoid::adjust_velocity()`

Located in libraries/AC_Avoidance/AC_Avoid.cpp.

This is the core function called by ModeLoiter, ModeAuto, etc.
Backing Up: It calculates if the vehicle is inside the safety margin (margin_cm) and generates a "Back Away" velocity vector.
Limiting: If the vehicle is approaching an obstacle, it clamps the velocity component parallel to the obstacle vector.

Source Code Reference

Proximity Core: libraries/AP_Proximity/AP_Proximity.cpp
Avoidance Core: libraries/AC_Avoidance/AC_Avoid.cpp

Practical Guide: Debugging Proximity Data

If avoidance isn't working, first check if the data is valid.

Open MAVLink Inspector: Look for DISTANCE_SENSOR messages.
Check PRX_LOG_RAW: Set this to 1 to log raw sensor values to the SD card.
The "Fence" Display: In Mission Planner/QGC, ensure the "Proximity" overlay is enabled. You should see a red boundary line appear around the drone when obstacles are detected.

How To: Setup a 360° Lidar (RPLidar)

360-degree Lidars are the "gold standard" for omnidirectional awareness. Here is how to integrate a common unit like the RPLidar A2/A3.

1. Physical Connection

Power: Most Lidars require a dedicated 5V BEC (they draw too much current for the Flight Controller's internal regulator).
Data: Connect TX (Lidar) to RX (FC) and RX (Lidar) to TX (FC) on a spare Serial port (e.g., SERIAL4).

2. Configuration

Set the following parameters in Mission Planner:

Parameter	Value	Description
SERIAL4_PROTOCOL	`11`	Set protocol to "Lidar360".
SERIAL4_BAUD	`115`	RPLidar typically runs at 115200 baud.
PRX1_TYPE	`4`	Select "RPLidar" driver.
PRX_ORIENT	`0`	Default orientation (0 = Forward). Adjust if you mounted it rotated.

3. Verification

Reboot the flight controller.
Open Mission Planner > Data Screen > Proximity.
You should see a "radar" display showing points around the vehicle.
Walk around the drone; the points should track your movement. If the points move with the drone when you rotate it, your compass or PRX_ORIENT is incorrect.