UAS.AMOGH.SITE
PolyUAS  ·  C-UASC Program  ·  2026 CYCLE ACTIVE

PolyUAS Cal Poly's autonomous UAS program: the aircraft, the autonomy, and the 2026 development plan

PolyUAS designs, builds, and flies autonomous aircraft for the CSU California Unmanned Aerial System Competition (C-UASC): autonomous flight and navigation, vision-based target identification and localization, and precision package delivery. This is the full program and the plan for the 2026 cycle, from airframe to onboard AI.

Competition
C-UASCAutonomy · vision · delivery
Flagship aircraft
12 kg quad12S · ~40 min · modular
Autonomy
Jetson + YOLOv8ArduPilot · MAVLink
2026 focus
Swarm + custom EECompute, power, radio, GCS
01The Competition

C-UASC: fly it, find it, deliver it

The CSU California Unmanned Aerial System Competition tasks student teams with designing, integrating, and demonstrating a UAS that flies autonomously and executes a set mission. Everything the program builds serves these three capabilities.

TASK 01

Autonomous flight & navigation

Waypoint autonomy, stable flight in wind and disturbance, and reliable state estimation. Owned by the flight controller and the autopilot tuning, not a human stick input.

TASK 02

Target ID, classify & localize

Onboard computer vision detects, classifies, and geo-locates targets from the gimbal camera, then hands coordinates to the flight stack. Real-time detection is the constraint.

TASK 03

Precision package delivery

An autonomous payload mechanism lowers and releases a package accurately at a target, energy-efficiently, without destabilizing the aircraft.

Faculty advisor: Dr. Xi Wu. Vehicles may be multi-copter, fixed-wing, or VTOL. PolyUAS fields multiple build teams (Osiris, Horus, and others) plus a Fall 2026 swarm initiative.

02The Program

Multiple aircraft, one autonomy stack, a shared avionics language

PolyUAS is not one drone. It is several build teams iterating airframes against the same competition, unified by a common avionics and autonomy stack (SIYI N7 autopilot, SIYI HM30 link, Jetson compute, ArduPilot + MAVLink). The 2026 cycle pushes on three fronts at once.

Flagship multicopter

Proven

The 7 kg quad platform (Osiris-class) that placed the team in C-UASC. The 2026 baseline, being refined, not replaced.

Fall 2026 swarm

In progress

Five to six aircraft flying as one coordinated swarm (Pixhawk 6C Mini per node). The software team is standing up the coordination nodes; new airframes are being spec'd around it.

Fixed-wing / VTOL

Exploration

Evaluating fixed-wing and VTOL platforms for longer range and endurance than a pure multirotor can reach. Powertrain trade studies underway.

03The Flagship Aircraft

A 7 kg autonomous quad, engineered end to end

The proven platform the 2026 work builds on. A modular multicopter with a vibration-isolated avionics bay, quick-detach arms, a custom energy-dense battery, and a full onboard autonomy stack. Specs from the team's C-UASC design review.

JETSON + CV SIYI N7 24V Li-ion PAYLOAD KV320 · 22" PROP5 kg THRUST eaSIYI A8 GIMBAL
Fig. 01 · Flagship quad subsystem layout (schematic)
Airframe~12 kg AUW heavy-lift quad · modular removable arms, protected avionics bay, packs into an IP67 case
StructureAluminum central plates + CF arms · blind-mate JX4 arm connectors (CAN + PWM), 5° motor dihedral
PropulsionSIYI E6 integrated propulsion · FOC ESC, PWM/CAN, ESC telemetry (RPM + fault), up to 6 kg/rotor
Power12S · two custom 6S4P 21700 packs · ~44 V, 18 Ah usable, ~27 A hover, ~40 min
Flight ctrlSIYI N7 (ArduPilot) · dual IMU + thermostatic control, RTL + geofence + lost-link failsafes
ComputeJetson Orin Nano · YOLO-nano + TensorRT INT8, >30 FPS; Rust/Zenoh stack
CameraSIYI A8 mini gimbal · forward, quick-disconnect mount
Link / GPSSIYI HM30 (5.8 GHz HD) · RTK GPS · video + MAVLink + control, ground control station
PayloadPassive spool + viscous damper + servo trap door · ~5 kg, delivery + drop, no active winch
04The 2026 Development Plan

What the team is building this cycle

Every workstream the program is pushing on, with an honest read on each: what is proven and being refined, what is a genuine build, and where ambition needs a reality check. Green tags mark work owned by the EE side.

Mechanical & airframe

Refine

The modular platform works. 2026 is a frame redesign around new interference-fit parts, plus continued refinement of the quick-detach arms, compliant landing gear, and the winch/hook/drop-box payload.

  • Frame redesign around the new IF part set
  • Payload winch + passive auto-hook + drop-box (energy-efficient release)

Propulsion & power

EE

The team builds its own energy-dense Li-ion packs (24 V, 22.5 Ah). Now they want to manage them: a custom BMS PCB with per-cell monitoring, protection, live telemetry, and smart charging.

  • Battery manufacturing: lock the cell config + assembly process
  • BMS + smart-charging PCB (EE): per-cell monitoring, balancing, protection, telemetry

Avionics & EE hardware

EE

The layer PolyUAS had no dedicated owner for. Moving from adapter-stacks and messy wiring to purpose-built boards the team explicitly wants.

  • Power-distribution PCB (EE): on-board stepdowns + clean routing to replace the wiring harness
  • Compute carrier + instrumentation (EE): isolated Orin power + peripheral breakout
  • Custom radio (EE): control + telemetry around proven RF silicon

Software & autonomy

Active

Upgrade the onboard brain from Jetson Nano to Orin Nano Super and keep the YOLOv8 detection pipeline running in real time, plus the swarm coordination nodes.

  • Compute upgrade: Jetson Nano → Orin Nano Super
  • Swarm nodes: coordinate 5-6 aircraft as one (in progress)
  • Thermal management (EE): keep the Orin at full clock, no throttling

Ground control station

Build

A purpose-built portable GCS: an embedded computer talking MAVLink to the aircraft over a radio, with a screen that maps and locates the drone in the field.

Scope it right: build the hardware (rugged unit, radio + power integration, and an antenna-tracker is the strong EE add-on). Run the proven software (QGroundControl / MAVSDK / pymavlink), don't rebuild maps and mission planning from scratch. Non-flight-critical, so a good, safe project.

Custom flight controller

Reality-check

An ambition on the wishlist: build the team's own flight controller.

Recommend against it as a flight article. An FC is safety-critical real-time control (IMU → EKF → rate loops → failsafes). A proven autopilot like the SIYI N7 running ArduPilot is a decade of flight-testing in one board; a homemade FC would be less safe and less capable after a year of work. Keep it a learning bench project only, or contribute to ArduPilot. Do not fly the competition on it.
05The EE Program

Own the power and thermal truth of the electronics

The electrical side PolyUAS is building from the ground up (Amogh Somisetty). The team already builds its own batteries and wants its own board-level electronics to match: a battery management system with live telemetry and smart charging, and custom PCBs that replace the messy wiring harness with clean, integrated power distribution. Plus a thermal-reliability layer for the onboard AI that the software team can't build.

◈ Edge-AI thermal reliability

Keep the AI running when it matters most

The Orin throttles under sustained vision inference in a hot canopy: clock drops, detection frame-rate drops, targets get missed. On a drone the rule is never sacrifice performance to stay cool, so the job is to keep it from throttling at all, with cooling and power headroom, not by reducing the AI.

Fig. 02 · Throttle onset: temp crosses limit, clock & FPS fall (illustrative)
  • AInstrument compute rails + junction temps under real load (software-only, no board)
  • BMap the thermal envelope: where and when it would throttle
  • CEngineer it out with cooling + power headroom so it holds full clock, wear accepted
◈ Smart battery (BMS)

A pack that reports and predicts its own health

The team builds its own energy-dense packs and wants its own BMS to match: per-cell monitoring, balancing, protection, live telemetry, and smart charging, all on one custom PCB. Today the pack flies unmanaged; this closes that gap.

BMS V · T · I balance · protect drifting cell flagged early
Fig. 03 · Per-cell sense → balancing, protection, live telemetry
per-cell Vmulti-point tempcurrentbalancingprotectionsmart chargingMAVLink telemetry

Flight rule: it warns and reports so the aircraft lands on its own terms. It never cuts power mid-flight for a recoverable condition, because losing the drone is worse than stressing the pack.

Avionics interconnect board

First board is designed: a passive Arm & CAN backplane (power-dist-v1) that consolidates the DroneCAN bus and the four arm interfaces onto one part. Schematic ERC-clean and verified against the SIYI N7 pinout. See it in 3D ↘ Passive interconnect first (near-zero risk), then stepdowns, conditioning, and accurate voltage/current sensing.

Custom radio board

Control + telemetry around proven RF silicon (SX1280 / SX126x). The SIYI HM30 carries HD video and control today; this would own a redundant control/telemetry link. Shielded from power/motor noise.

Power architecture

Isolate avionics rails from propulsion so motor noise never browns out the compute. Dedicated BECs + bulk caps, the number-one companion-computer failure fix.

06Hardware · Board

The Arm & CAN backplane, in 3D

The first custom board: a passive backplane that consolidates the aircraft's DroneCAN bus and the four arm interfaces onto one part. Drag to rotate, scroll to zoom, look from any angle. This is a preliminary model (placeholder outline, pending the avionics-tray CAD); the wiring it captures is verified against the team's design docs.

◈ power-dist-v1 · passive

What it does

Ingests 48 V from the PDB, the SIYI N7 CAN bus, and N7 PWM 1-4; lands them on four clean JX4 arm connectors plus a Remote-ID CAN drop, with 120 Ω bus termination. Four independent 48 V pass-throughs mean no single trace carries the full motor current.

  • Schematic captured, ERC-clean (0 violations), netlist verified against the N7 pinout
  • Footprints assigned (arm connector = placeholder pending the SIYI JX4 footprint)
  • Next: tray-CAD outline, real JX4 footprint, route and fabricate

Preliminary placement on a 120 × 90 mm placeholder outline. Verified against the SIYI N7 pinout and the Osiris wiring diagram. XT30 bodies not shown (no stock 3D model).

07Power Architecture

Compute never shares a rail with the motors

The single most important electrical decision on the aircraft. Motor switching dumps noise on any shared rail; the number-one way a companion computer reboots mid-flight is drinking from the same supply as the ESCs.

12S ~44V 2× 6S4P + BMS PDB SIYI E6 ×4 MOTOR · KV320MOTOR · KV320MOTOR · KV320MOTOR · KV320 ◄ isolation barrier · dedicated regulators + bulk caps ► AVIONICS BEC BANK 5V clean · 5V compute SIYI N7Jetson OrinSIYI A8 gimbalRadio link propulsion (noisy) avionics (clean)
Fig. 04 · Isolated power tree
08Roadmap

Fly on proven parts, layer the custom work in

The aircraft flies today on a proven stack. Custom hardware layers on without ever putting a competition season hostage to a board respin. Boards clear a review gate before they cost money or fly.

Now

Baseline flying

Proven quad on commercial avionics. Frame + battery redesign underway.

Phase 1 · now

Edge-AI thermal

Software-only Orin characterization + detector. No board needed.

Phase 2 · now

Interconnect + BMS

Arm & CAN backplane (schematic done, ERC-clean), then the smart battery board.

Phase 3

Radio + GCS

Custom control link + the portable ground station.

Phase 4

Swarm

Multi-aircraft coordination integrated with the stack.

Phase 5

Flight → C-UASC

Integration, flight test, competition.

Guiding principle: build around the mature open cores (ArduPilot, MAVLink, QGroundControl, proven RF and BMS silicon). Innovate at the edges (power, thermal, integration, radio hardware, autonomy). Reinventing a safety-critical core is where student teams lose seasons.