Orion — ROS2 Quadruped Robot

A custom-built 12-DOF quadruped robot designed, built, and programmed from scratch over 8 months. Uses an NVIDIA Jetson Orin Nano for high-level AI and perception, and a custom STM32 PCB for real-time low-level control.

Check out the Demo Video!

Key Features

• ROS2 / AI — Isaac ROS Docker on Jetson Orin Nano, TensorRT for accelerated inference
• Perception — Stereo depth (MIDAS, SGBM) via dual CSI cameras + MediaPipe hand-pose recognition
• SLAM & Autonomy — RPLIDAR A1M8 with slam_toolbox and nav2 for mapping and point-and-click pure-pursuit navigation
• Embedded Control — Custom STM32 FreeRTOS firmware: IK, walking gaits, IMU-based stabilization
• Custom PCBs — Multi-layer SMD boards for power distribution, servo actuation, and sensor interfacing
• RL / Sim — PPO training (RSL RL) in NVIDIA Isaac Lab, flat and rough terrain environments
• Robot Eyes — Expressive OLED eye animations via a custom Python/OpenCV node

---

Software Architecture

Jetson — Isaac ROS Workspace

Package	Description
orion_base	STM32 UART bridge, joystick parser, command mux, robot eyes, joint state republisher
orion_camera	Dual CSI GStreamer node, MIDAS/SGBM depth, MediaPipe hand pose
orion_lidar	RPLIDAR A1M8 driver, slam_toolbox
orion_navigation	nav2, Pure-Pursuit path following
orion_bringup	Launch files, Docker config, systemd auto-start
orion_msgs	Custom messages: motion cmd, eye cmd, battery voltage, IMU, leg info
orion_urdf	URDF model with SolidWorks meshes, RViz launch

STM32 · Orion Firmware

• 5 FreeRTOS tasks: Real-time IK & gait control (Realtime) with 3-DOF per leg and 6-DOF body posture control, UART RX/TX with Jetson via DMA (High/Normal), IMU reading, Battery monitoring.

Reinforcement Learning

• PPO training in Isaac Lab with flat and rough terrain environments.
• SolidWorks → URDF → USD pipeline.

---

Software Gallery

LiDAR SLAM	Autonomous Navigation
Original Camera Feed	TensorRT Depth Map
Hand Pose Detection	Reinforcement Learning in Isaac Lab

Foxglove Dashboard

---

Electrical & Electronics

Custom multi-layer KiCad PCBs, fully SMD-assembled. Multiple I2C buses with LTC4311 active termination, extensive test pads and debug headers.

• Control Board — STM32F401, BNO055 IMU, PCA9685 servo driver
• Power Board — Battery distribution, INA3221 voltage monitoring (3 rails), high-current servo lines, battery alarm

Electrical Gallery

Control Board	Power Board

Control Board	Control Board Compute	Power Board

---

Mechanical Design

Chassis iteratively modelled in SolidWorks. 3D printed with 2-part silicone injection-molded feet. Bearings in all 12 joints. Magnetic mounts for tool-free access to internal electronics. Single-leg linear-rail test jig used to validate IK and gait before full integration.

Main body parts	Leg Assembly & Silicone Feet
Inside Orion (Jetson + PCBs + Display only)	Inside Orion (including Batteries + Lidar + Cameras)

---

Repository Structure

Important directories:

Quadruped/
├── Software/
│   ├── Jetson/workspace/isaac_ros-dev/   # Isaac ROS Docker workspace (ROS2 packages)
│   ├── STM32Firmware/
│   │   ├── Orion/                        # Main FreeRTOS firmware (CMake + STM32CubeIDE)
│   │   └── Orion-Controls/               # Legacy PlatformIO firmware
│   ├── reinforcement-learning/           # Isaac Lab PPO training
│   ├── TestScripts/                      # Servo calibration, camera, depth mapping tests
├── models/                               # SolidWorks parts & assemblies, STL, 3MF exports → README
├── urdf/                                 # URDF iterations (ROS2, RL variants) → README
├── pcb/                                  # KiCad projects (Control Board, Power Board) → README