Advanced Autonomous Drone Solution
Advanced Autonomous Drone Solution
One Chip. From Perception to Propeller.
OneKiwi's Advanced Autonomous Drone Solution integrates edge AI perception, autonomous navigation, hard-real-time PX4 flight control, deterministic I/O and motor safety on a single Renesas RZ/V2H processor — replacing the AI companion computer, flight controller and deterministic motor-safety controller found in conventional split architectures.
One processor instead of three boards
Multiple computing boards mean more weight, more power, more cabling and more parts to integrate and qualify. The RZ/V2H puts each role in its own hardware-isolated domain instead.
Conventional Split Architecture → OneKiwi RZ/V2H
| AI companion computer | Cortex-A55 cluster + DRP-AI3 |
| PX4 flight controller | Dedicated Cortex-R8 real-time core |
| I/O co-processor | Second Cortex-R8 core |
| ESC & safety controller | Dedicated Cortex-M33 |
| Board-to-board control link | On-chip shared-memory IPC |
| Multiple computing boards | One integrated processing platform |
AI advises. PX4 authorizes. M33 enforces.
The Linux/AI domain never touches the ESCs directly — a perception error or application crash cannot bypass flight authority or the motor-safety gate.
AI proposes
Linux + AI analyze camera and navigation data, propose mission or trajectory actions.
PX4 validates
Checks flight mode, estimator health, geofence, pilot input and failsafe state.
M33 enforces
Verifies the command stream before generating physical PWM/DShot output.
Heterogeneous computing on one processor
The RZ/V2H combines high-performance application processing, AI acceleration and real-time control domains on a single device.
| 4× Cortex-A55, up to 1.8 GHz | Autonomous navigation, mission planning, VIO/SLAM, video and payload services — Yocto Linux + ROS 2 |
| DRP-AI3 accelerator | Object detection, tracking and vision inference — Renesas DRP-AI software stack |
| Cortex-R8 Core 0, up to 800 MHz | PX4 flight management, estimation and control loops — NuttX + PX4 |
| Cortex-R8 Core 1, up to 800 MHz | Sensor I/O, CAN-FD, RC, battery and telemetry services — NuttX + PX4 I/O |
| Cortex-M33 | PWM/DShot generation, watchdog, arming logic and motor cutoff — dedicated deterministic runtime |
Integrated AI acceleration. 8 TOPS dense / up to 80 TOPS sparse AI performance, up to 10 TOPS/W efficiency, hardware-accelerated computer vision without a discrete GPU — supporting object detection, tracking, segmentation and visual inspection workloads. Passive thermal implementation depends on the final enclosure, workload and aircraft operating environment.
Flight control that never waits on Linux
PX4 flight-control workloads run on dedicated Cortex-R8 cores, physically separated from Linux and AI processing that can create variable CPU and memory load.
| Rate control | 1,000 Hz |
| Attitude control | 500 Hz |
| Position control | 250 Hz |
| Actuator-setpoint transfer | 400 Hz, CRC32-protected |
| Vehicle-state publication to A55 | 100 Hz |
| IMU sampling | Up to 2 kHz |
Built for advanced airframes
The Cortex-M33 safety domain supports deterministic PWM and bidirectional DShot generation for 4–8 ESC channels. Supported airframe classes include quadcopter, hexacopter, X6 coaxial hexacopter, octocopter and custom multirotor configurations. Bidirectional DShot provides digital checksum-protected motor commands, no analog throttle calibration, ESC telemetry and RPM feedback, and dynamic RPM-based notch filtering for improved vibration suppression.
Four layers, validated at the system level
A single MPU remains a common physical point of failure — redundancy comes from the validated system: sensors, navigation sources, links and airframe.
Sensor
Primary and backup IMUs, GNSS, barometer, magnetometer, VIO or optical-flow positioning.
Navigation
Estimator-health-based source selection — not simply the presence or absence of GNSS.
Communication
Primary and secondary telemetry paths; link degradation is reported to PX4 but cannot bypass flight authority.
Airframe
X6 hexacopter reference — motor-out fault tolerance not achievable with a standard quadcopter.
Mission intelligence without disturbing flight control
The A55 and DRP-AI3 domain executes perception and mission workloads — object detection and tracking, segmentation and inspection, mission planning, VIO/SLAM, optical-flow navigation, GNSS-denied positioning, AI-assisted gimbal framing, video encoding and evidence recording — without disturbing the real-time flight-control loops.
The solution supports visible-light, EO/IR and compatible three-axis gimbal payloads through MAVLink Gimbal Protocol v2. Command priority is strictly separated: safety and failsafe functions first, then pilot or ground-control command, then active mission policy, then AI tracking proposal.
Authority boundary. AI results remain advisory. PX4 retains authority over navigation mode, command acceptance and failsafe selection.
Where one-chip autonomy pays off
Industrial Inspection
Power and utility assets, solar panels, oil and gas facilities, industrial plants, bridges, construction monitoring.
GNSS-Denied Operations
Warehouses, factories, tunnels, underground facilities, indoor asset inspection, weak-GNSS infrastructure.
Delivery & Logistics
Last-mile delivery, industrial material transport, medical and emergency delivery, visual delivery confirmation.
Public Safety & Mapping
Search and rescue, emergency assessment, fire and disaster monitoring, security patrol, surveying and 3D mapping.
Claims boundary. Publish as “fault-contained, fail-safe and redundancy-ready” — never as an unqualified “failure-redundant single chip.” A single MPU is still a common physical point of failure; aircraft-level redundancy requires validated sensor, power, navigation, communication and airframe redundancy on top, followed by system-level validation.