GannFlight

Genetic Neural Network Control Framework for ArduCopter (Multi-Instance SITL)

Overview

GannFlight is a research framework for launching and coordinating multiple ArduCopter SITL (Software-In-The-Loop) instances during controller experiments.

This repository now includes two command-line scripts designed to work together:

• controller.py: runs on the machine that has ArduPilot installed and manages SITL processes.
• client.py: sends start/stop/list commands to controller.py over TCP.

---

Repository Organization

• controller.py - TCP server that starts/stops SITL instances.
• client.py - CLI client for controlling instances remotely or locally.
• main/simulation_controller.py - backward-compatible entrypoint that forwards to controller.py.
• main/client.ipynb - notebook used for experimentation/training workflows.
• images/ - plots and output images from simulation/training runs.

---

Requirements

• Linux (tested on Ubuntu)
• Python 3.10+
• ArduPilot source checkout (default expected path: /home/robotics/ardupilot)

Recommended: create and use a Python virtual environment for your experiment tooling.

---

Script Reference

controller.py

Starts a TCP server and accepts these commands:

• start <instance_id> <out_port>
• stop <instance_id>
• list
• stop-all

Controller options

python3 controller.py --help

Important arguments:

• --host (default 0.0.0.0): interface to bind for incoming client connections.
• --port (default 14500): TCP port for control commands.
• --ardupilot-path (default /home/robotics/ardupilot): local ArduPilot checkout path.
• --out-host (default 127.0.0.1): destination host for SITL MAVLink --out stream.

client.py

Sends one command per invocation.

python3 client.py --help

Subcommands:

• start <instance_id> <out_port>
• stop <instance_id>
• list
• stop-all
• raw "<command>"

---

Usage: Single Machine

This mode is for when controller + client + MAVLink consumer all run on one computer.

1) Start the controller

python3 controller.py \
  --host 0.0.0.0 \
  --port 14500 \
  --ardupilot-path /home/robotics/ardupilot \
  --out-host 127.0.0.1

2) Start one or more SITL instances from another terminal

python3 client.py --host 127.0.0.1 --port 14500 start 0 14550
python3 client.py --host 127.0.0.1 --port 14500 start 1 14551
python3 client.py --host 127.0.0.1 --port 14500 list

3) Connect your MAVLink consumer

For instance 0 started with out_port=14550, connect your consumer to UDP port 14550 on the same machine (example pymavlink endpoint: udpin:0.0.0.0:14550).

4) Stop instances

python3 client.py --host 127.0.0.1 --port 14500 stop 0
python3 client.py --host 127.0.0.1 --port 14500 stop-all

---

Usage: Multiple Machines

This mode is for when SITL runs on one machine and your experiment/training code runs on another.

Topology

• Machine A (simulation host): runs controller.py and ArduPilot SITL.
• Machine B (experiment host): runs client.py and your training/analysis code.

Assume:

• Machine A IP: 192.168.1.124
• Machine B IP: 192.168.1.40
• Control port: 14500

1) On Machine A, start controller with --out-host set to Machine B

python3 controller.py \
  --host 0.0.0.0 \
  --port 14500 \
  --ardupilot-path /home/robotics/ardupilot \
  --out-host 192.168.1.40

This ensures each SITL instance is launched with --out 192.168.1.40:<out_port>.

2) On Machine B, send commands to Machine A

python3 client.py --host 192.168.1.124 --port 14500 start 0 14550
python3 client.py --host 192.168.1.124 --port 14500 start 1 14551
python3 client.py --host 192.168.1.124 --port 14500 list

3) On Machine B, listen on those UDP ports

Your consumer/training loop should bind to each configured out_port (for example 14550, 14551).

4) Networking checklist

• Open TCP 14500 from Machine B to Machine A.
• Open UDP 14550+ (or your chosen range) to Machine B.
• Ensure no local firewall blocks these ports.
• Verify routing between subnets/VLANs if machines are not on the same LAN.

5) Shutdown

python3 client.py --host 192.168.1.124 --port 14500 stop-all

---

Notes

• Instance IDs should be unique per controller process.
• Keep out_port unique per instance to avoid UDP conflicts.
• Use list to inspect currently managed instances.
• Use main/simulation_controller.py only if older automation still points to that path.

---

Example Output