PATHIKA: Autonomous Ground Vehicle

The vehicle

PATHIKA is a 4×4 skid-steer unmanned ground vehicle built for the TEKNOFEST 2026 Unmanned Ground Vehicle competition. It has to complete the course in two runs: one remote-controlled and one fully autonomous.

Component	Model	Notes
Main computer	NVIDIA Jetson Orin NX 16 GB	100 TOPS, CUDA, TensorRT
LiDAR	Slamtec RPLidar A1M8	360°, 12 m range
Depth camera	Intel RealSense D455	87° FOV, built-in IMU
Drive camera	Logitech C920S	1920 × 1080
Aiming camera	Feelworld 4K10X	10× optical zoom
Motors / ESCs	Flipsky 6384 × 4 / FSESC 6.9 × 4	BLDC, FOC over CAN, 6 kW total
Battery	108 × 18650 Li-ion (12S9P)	44 V nominal, 1.1 kWh
Safety MCU	STM32 NUCLEO-G0B1RE	independent emergency stop & watchdog

Software architecture

Everything runs on ROS 2 Humble as a set of packages, developed and tested in Gazebo first:

• pathika_localization: EKF sensor fusion with robot_localization (wheel odometry + IMU) and slam_toolbox mapping.
• pathika_navigation: Nav2 with the MPPI controller, plus a dedicated slalom planner for the cone section.
• pathika_perception: LiDAR-based cone detection publishing cone poses, with YOLOv8 visual detection on the camera stream (in progress).
• pathika_control: skid-steer kinematics that turn /cmd_vel into four wheel velocities, and PID motor control through VESC over CAN.
• pathika_bringup: top-level launch files and the finite-state-machine mission manager that switches modes and stages on /mission_cmd and reports on /mission_status.
• pathika_description / pathika_msgs: the URDF/xacro model (chassis, wheels, LiDAR, IMU, cameras) and custom messages (Cone, Detection, TurretCommand, MissionStatus).

A key design choice is that safety does not depend on the main computer: the STM32 supervises an independent emergency stop and watchdog, so a crashed node or a hung Jetson cannot leave the motors running.

My part

In a software team of three, I own SLAM, the slalom planner, the control layer, the FSM mission manager and the bringup, and co-own the vehicle description. Perception, EKF fusion and Nav2 configuration are owned by my teammates.

# full system in simulation: Gazebo + SLAM + Nav2 + perception + control + FSM
ros2 launch pathika_bringup bringup.launch.py use_sim_time:=true

# drive the mission manager
ros2 topic pub /mission_cmd std_msgs/msg/String "data: 'set_autonomous'"
ros2 topic echo /mission_status

Status

The stack runs end to end in simulation, and the work now is bringing it onto the real vehicle: sensor calibration, controller tuning on the real drivetrain, and the perception models that are still being trained. I will add measured results from the field once the competition runs are complete.