UOA FSAE — Autonomous System

ROS 2 workspace for the University of Auckland Formula SAE autonomous go-kart. Currently, the car drives itself around a cone-lined track using a ZED stereo camera for perception, Kalman-filter-based landmark mapping, fast-tube path planning, and Stanley lateral control, all bridged to the physical kart over CAN.

Platform: NVIDIA Jetson (L4T 35.4) · ROS 2 Jazzy (on the autonomous PC, Humble on the Jetson - needs to be upgraded) · ZED 2i stereo camera Build system: colcon (ament_cmake + ament_python)

---

Repository Layout

autonomous/
├── src/
│   ├── fsae_interfaces/        # Shared ROS msg/srv definitions (ament_cmake)
│   ├── fsae_control/           # Stanley lateral controller (ament_python)
│   ├── fsae_planning/          # Path planning — fast-tube & centerline (ament_python)
│   ├── fsae_perception/
│   │   ├── zed_perception/     #   ZED camera + YOLO cone detection (C++, TensorRT)
│   │   └── fsae_slam/          #   Cone landmark SLAM (C++, skeleton — see below)
│   ├── gocart/
│   │   ├── gocart_bringup/     #   Main launch files (autonomous, scrutineering)
│   │   ├── gocart_control/     #   Ackermann-to-CAN bridge, teleop, mock stimulus
│   │   ├── gocart_driver/      #   CAN decoder for Jetson Nano
│   │   ├── gocart_description/ #   URDF model + meshes
│   │   └── CanTalk/            #   Serial CAN adapter driver (submodule)
│   ├── scrutineering/          # FSAE technical inspection mission controller
│   ├── visualisation/
│   │   ├── base_tf/            #   Static TF broadcaster
│   │   ├── cone_map_foxglove_visualiser/
│   │   ├── path_planning_visualiser/
│   │   └── pure_pursuit_visualiser/
│   └── third_party/
│       ├── foxglove-bridge/    #   WebSocket bridge for Foxglove Studio
│       └── yolov7/             #   YOLO model training reference
├── Formula-Student-Driverless-Simulator/   # FS Driverless Sim (submodule)
├── .devcontainer/              # Docker dev environment (Jetson + ZED containers)
├── makefile                    # Dev environment build helper
├── mock_params.yaml            # Parameter file for mock controller testing
└── requirements/               # pip dependency lists

---

Package Reference

Core Pipeline

Package	Type	Description	Executables
fsae_interfaces	cmake	All custom .msg and .srv definitions	(msg/srv generation only)
zed_perception	cmake	ZED camera node: stereo capture → YOLO (TensorRT) → cone detections + localisation	zed_launch_node, cone_subscriber, localisation_subscriber
fsae_slam	cmake	Cone landmark SLAM (C++ skeleton — porting from Python reference)	cone_landmark_mapper
fsae_planning	python	Path planning algorithms	fasttube, fasttube_without_kalman, centerline_planner, center_line
fsae_control	python	Stanley lateral controller	controller

Go-Kart Hardware

Package	Type	Description	Executables
CanTalk	python	Serial CAN adapter driver (candapter)	candapter_node
gocart_control	python	Ackermann-to-CAN, system status, teleop, mock stimulus	ack_to_can_node, as_status_node, trajectory_follower, joystick_teleop, mock_stimulus
gocart_driver	python	Jetson Nano CAN bus decoder	can_decoder_jnano
gocart_description	python	URDF model and STL meshes	(data only)
gocart_bringup	python	Launch files for the full autonomous system	(launch only)

Events & Visualisation

Package	Type	Description	Executables
scrutineering	python	FSAE technical inspection mission	inspection_mission_node, test_node
base_tf	python	Static TF broadcaster for kart frame	base_tf
cone_map_foxglove_visualiser	python	Publishes cone map markers for Foxglove	visualiser
path_planning_visualiser	python	Trajectory & image throttle visualisation	visualiser, visualiser2, image_throttler
pure_pursuit_visualiser	python	Pure pursuit debug visualisation	visualiser

---

Message Definitions (fsae_interfaces)

Messages (17)

Message	Key Fields
Detections	Pose car_pose, Point[] yellow, Point[] blue, Point[] small_orange, Point[] big_orange
Track	Point[] cones
Cone	uint32 id, float32 confidence, uint8 colour, PoseWithCovariance pose
ConeMap / ConeMapStamped	Cone[] cones (± header)
ConeStamped	Header, Cone
CAN / CANStamped	uint16 id, bool is_rtr, uint8[8] data (± header)
HardwareStates / HardwareStatesStamped	EBS, TS, gear, master switch, ASB, brakes (± header)
MissionStates / MissionStatesStamped	mission_selected, mission_finished (± header)
AllStates	uint16 id, AckermannDrive[] states
AllTrajectories	uint16 id, PoseArray[] trajectories
BoundaryStamped	Header, float32[] coords
OccupancyGrid	Header, uint8[] occupancy_grid, width, height, resolution
Pulse	bool value, string event_type

Services (1)

Service	Request	Response
CANSendReq	CAN can	bool sent

---

Launch Files

gocart_autonomous.launch.py — Full Autonomous Pipeline

The primary launch for competition runs. Namespace: moa.

ZED Camera → Cone Landmark Mapper → Fast-Tube Planner → Stanley Controller → CAN Adapter → Ackermann-to-CAN

#	Package	Executable	Node Name	Key Params
1	zed_perception	zed_launch_node	perception	—
2	fsae_slam	cone_landmark_mapper	cone_landmark_mapper	invert_cones: true
3	fsae_planning	fasttube	centerline_planner	invert_cones: true
4	fsae_control	controller	stanley_controller	vel: 9.0
5	CanTalk	candapter_node	—	—
6	gocart_control	ack_to_can_node	ack_to_can	can_id: 0x300

Optional (set visualize = True): image_throttler + foxglove_bridge on port 8765.

gocart_autonomous_no_kalman.launch.py — Without Kalman Filter for cone positions

Same pipeline but skips cone_landmark_mapper, uses fasttube_without_kalman instead, runs at vel: 10.0, and always enables Foxglove visualisation.

Other Launch Files

Launch File	Location	Purpose
base.launch.py	gocart_bringup	Base nodes: ack_to_can + sys_status + candapter
scrutineering.launch.py	gocart_bringup	base.launch + inspection mission node
camera_stimulus_test_launch.py	gocart_control	Full pipeline with mock CAN stimulus (bench testing)
motec_mock_controller_test_launch.py	gocart_control	CAN adapter + mock stimulus only
gocart_driver.launch.py	gocart_driver	CAN decoder node
fsae_slam.launch.py	fsae_slam	SLAM node standalone
zed_perception.launch.py	zed_perception	ZED camera node standalone
path_planning.launch.py	fsae_planning	Dynamic planner selection via node_name arg
pure_pursuit_visualiser_launch.py	pure_pursuit_visualiser	Pure pursuit visualisation
urdf_model.py	gocart_description	Spawn URDF in Gazebo

---

Architecture

                    ┌──────────────┐
                    │  ZED Camera  │
                    │  (stereo +   │
                    │   YOLO)      │
                    └──────┬───────┘
                           │ /cone_detection (Detections)
                           │ /car_position   (Pose)
                           ▼
                    ┌──────────────┐
                    │  fsae_slam   │
                    │  (landmark   │
                    │   mapper)    │
                    └──────┬───────┘
                           │ /left_track  (Track)
                           │ /right_track (Track)
                           ▼
                    ┌──────────────┐
                    │ fsae_planning│
                    │  (fast-tube) │
                    └──────┬───────┘
                           │ /selected_trajectory
                           ▼
                    ┌──────────────┐
                    │ fsae_control │
                    │  (Stanley)   │
                    └──────┬───────┘
                           │ /cmd_vel (AckermannDrive)
                           ▼
              ┌────────────┴────────────┐
              │    gocart_control       │
              │  (ack_to_can_node)      │
              └────────────┬────────────┘
                           │ /can (CAN)
                           ▼
              ┌────────────┴────────────┐
              │       CanTalk           │
              │   (candapter_node)      │
              └────────────┬────────────┘
                           │ Serial USB
                           ▼
                     Physical Kart

---

Building

# Source ROS 2 Jazzy
source /opt/ros/jazzy/setup.bash

# Build all packages (skip those needing ZED SDK / external deps)
colcon build --packages-skip zed_perception foxglove_bridge

# Source the workspace
source install/setup.bash

Note: zed_perception requires the ZED SDK 4, CUDA, and TensorRT — build it on the Jetson or a machine with the SDK installed. foxglove_bridge is a third-party submodule that builds separately.

---

Running

# Full autonomous pipeline
ros2 launch gocart_bringup gocart_autonomous.launch.py

# Without SLAM (direct cone-to-path, with visualisation)
ros2 launch gocart_bringup gocart_autonomous_no_kalman.launch.py

# Bench test with mock CAN stimulus
ros2 launch gocart_control camera_stimulus_test_launch.py

# Scrutineering inspection
ros2 launch gocart_bringup scrutineering.launch.py

---

Submodules

Submodule	Path	Source
CanTalk	src/gocart/CanTalk	UOA-FSAE/CanTalk
foxglove-bridge	src/third_party/foxglove-bridge	foxglove/ros-foxglove-bridge
yolov7	src/third_party/yolov7	WongKinYiu/yolov7
FS Driverless Simulator	Formula-Student-Driverless-Simulator	FS-Driverless/Formula-Student-Driverless-Simulator

---