Orion

Orion is a 12-DOF quadruped designed, built and programmed from scratch over eight months by University of Waterloo mechatronics student Ashish Agrahari. The compute stack splits between an NVIDIA Jetson Orin Nano — running ROS 2 Humble in Isaac ROS Docker containers with TensorRT-accelerated stereo depth (SGBM, MiDaS), MediaPipe hand-pose recognition, RPLIDAR A1M8 SLAM via slam_toolbox and nav2 point-and-click navigation — and a custom STM32 PCB running FreeRTOS firmware for real-time IK, gait generation and IMU stabilization. Locomotion policies were trained with PPO in NVIDIA Isaac Lab over flat and rough terrain, following a SolidWorks to URDF to USD pipeline. Custom multi-layer KiCad PCBs (STM32F401 control board, power distribution board), 3D-printed chassis with silicone-molded feet, Foxglove dashboard and expressive OLED eyes round out the build. Documented end-to-end on GitHub. Verification level: limited — key specs are maker-reported and worth confirming directly.

Also featured on NVIDIA's developer forums (Sept 2026) and Circuit Digest. 8-month solo build; custom KiCad PCBs; SolidWorks to URDF to USD pipeline.