Project Devlog: Building the Guardian X-1 Hybrid Autonomous Platform
Date: August 2026
Status: Phase 3 — Structural Assembly, Power Distribution & Edge-AI Integration
Welcome back to the build log for the Guardian X-1. Over the past few weeks, we transitioned straight from CAD models and stress simulations to the workbench, building a high-performance, hybrid autonomous rover-drone capable of dual-mode mobility backed by a local neural processing unit (NPU) stack.
- Mechanical Fabrication & Modular Chassis:
Fabricating a vehicle designed to handle rotor harmonics and ground-drive torsion required a hybrid carbon-fiber mainplate coupled with high-tolerance 3D-printed enclosures: CF-PETG for general housing and PA6-CF (Carbon-Filled Polyamide) for motor boom arms and wheel struts.
1.Thermal Shrinkage: Printing structural parts in PA6-CF demanded a fully enclosed, heated chamber. Dialing the bed to 105°C and chamber ambient to 50°C eliminated corner warping
2.Fit-Tolerance Testing: We integrated a tight 0.2mm fit tolerance for interlocking drawer assemblies. The sliding utility deck now seats smoothly into the lower electronics floor without binding.
3.Hardware Assembly: Every structural joint uses heat-set brass threaded inserts paired with M3 socket-head cap screws, eliminating stripped threads during torque tests.
Power Architecture & Smoke Testing
Wired dual CNHL 4S 2200mAh 100C LiPo packs to feed the flight stack and companion compute tier.
1.The Smoke Scare: A brief ground plane bridge on an auxiliary harness popped the inline fuse on the SpeedyBee F405 V4 stack instantly, saving the Raspberry Pi 5. Always use a smoke stopper on first power-ups.
2.Current Draw: Idling under a ROS 2 launch payload draws roughly 1.4A at 16.8V approx. 23.5W. Full motor bench simulations sit comfortably within safe headroom for our 55A ESC tier.
- Compute Stack & Edge-AI Bring-Up
Flashing an optimized Ubuntu Server image onto the NVMe SSD via the M.2 HAT on the Raspberry Pi 5 allowed us to boot our ROS 2 workspace:
[INFO] [launch]: All nodes executed successfully…
[Node-1] /rplidar_node: spinning at 5.5 Hz, 360-degree scan active.
[Node-2] /hailo_yolo_node: NPU loaded successfully. Inference latency: 11.2ms.
[Node-3] /depth_pointcloud_node: Arducam ToF publishing depth streams.
Thermal Throttling Fix: The Raspberry Pi AI Kit (Hailo-8L NPU) delivers 13 TOPS for real-time YOLO and Face ID but generates intense heat. We milled a custom aluminum heatsink bridging the Pi 5 SoC and Hailo stack, paired with a 30mm variable-speed blower fan. Temperatures under full load dropped from 82°C to a stable 54°C.
-
Bill of Materials StatusSubsystemComponentStatusEst. CostCompute & AIRaspberry Pi 5 (8GB) + Hailo-8L NPUIntegrated & Tested$150StorageM.2 HAT + 128GB NVMe SSDFlashed & Booted$45Flight StackSpeedyBee F405 V4 55A StackWired & Bench-Tested$75Aerial Propulsion4x EMAX ECO II 2207 MotorsMounted to PA6-CF Booms$56Ground Drivetrain4x N20 Gearmotors + WheelsAssembled in Struts$28Chassis FrameHybrid Carbon + CF-PETG EnclosuresFully Assembled$50Power & Energy2x CNHL 4S 2200mAh LiPo + UBECCommissioned & Fused$65Mapping & SensorsRPLIDAR, ToF, Thermal, Radar, RGBCommunicating via ROS 2$332Hardware / WiringFasteners, Inserts, HarnessesInstalled$20TOTAL$847
-
Next Steps
1.Odometry Calibration: Sync N20 wheel encoder data with LiDAR scan matching for indoor dead-reckoning.
2.PID Tuning: Write the state machine controller for ground traction vs. vertical lift transitions.
3.Field Testing: Maiden tethered ground run in an open paved lot.
