Vinayak Nair / AEROSPACE

Home / Projects / PRJ-08 KESTREL

Archived Reported CDR readiness No flight campaign

Project KESTREL

THROW-LAUNCH UAV · VISION-ONLY TERMINAL GUIDANCE · SUB-5 cm CEP TARGET

A hand-thrown autonomous UAV concept targeting recovery to a marked pad with sub-5 cm circular error probable using vision-only terminal guidance. The closeout record reached CDR readiness, but no hardware qualification or flight campaign followed. This page separates design intent from evidence.

Illustration of the KESTREL hand-launched UAV arc toward a marked landing pad, stamped archived.

00Maturity & evidence ledger

Archived Design-stage No flight test No public repository

Claim boundary: KESTREL produced a requirements and architecture record that was closed out at reported CDR readiness. It did not produce a qualified airframe, flight-test dataset, or demonstrated landing-accuracy result.

Evidence source: programme closeout narrative. No source repository, CDR package, bench log, or flight artifact is publicly linked.

Evidence layerRecorded outcomePublic artifactClaim allowed
RequirementsTargets and intended verification methods definedPortfolio summary onlyDesign intent
ArchitectureVehicle, avionics, and terminal-guidance concept recordedNo public design packageDesign approach
Readiness reviewCloseout record states CDR readinessNo public board recordReported maturity
Hardware qualificationNot completedNoneNo qualification claim
Flight performanceNot testedNoneNo accuracy claim

01Problem

A small UAV that can be thrown by hand and recovered to a precise, pre-marked point without GNSS has obvious utility: no launch equipment, no runway, no dependence on a satellite constellation that may not be available. The difficulty is entirely in the terminal phase. GNSS-quality position is metres; the requirement here was centimetres, which means the last few seconds have to be flown on vision.

Problem statement. Recover a hand-launched UAV to a marked landing pad with sub-5 cm CEP, using vision as the sole terminal position reference.

02Requirements

IDRequirementValueVerification
KS-001Landing accuracy shall meet the precision target< 5 cm CEPFlight test campaign
KS-002Launch shall require no ground equipmentHand throwDemonstration
KS-003Terminal guidance shall not depend on GNSSVision onlyInspection + test
KS-004Transition from launch to stable flight shall be autonomousThrow detectionFlight test
KS-005Autopilot shall be a flight-proven platformPixhawk 6CSelection review

03Design approach

The recorded design specified a vision-only extended Kalman filter as the intended terminal position reference, fusing pad-relative visual measurements with inertial propagation on a Pixhawk 6C flight controller. The launch transient, a hand throw imparts a large, poorly characterised initial rate, was treated as a distinct flight phase with its own detection logic and recovery mode rather than as an edge case of normal flight.

AutopilotPixhawk 6C
Terminal navigationVision-only EKF, pad-relative
Launch modeThrow detection with autonomous stabilisation recovery
Accuracy target< 5 cm CEP at touchdown
Maturity reachedCDR readiness

04Status at closeout

  • The closeout narrative records a requirements set with verification methods assigned; the source package is not public.
  • The recorded design specified a vehicle configuration and avionics architecture; no public review package is available.
  • The narrative describes a vision-only terminal-guidance concept; no public analysis artifact is available.
  • The closeout record reports CDR readiness; no public board disposition is available.
  • No flight test campaign executed, the programme closed before hardware qualification.
Target, not resultFlight evidence: none

The sub-5 cm CEP figure is a design target that was never flight-demonstrated. It is reported here as a requirement, not as a result, and it should not be read as one.

05Why it stopped

KESTREL was archived because its most valuable technical content, vision-only terminal guidance in the absence of GNSS, was a strict subset of a more general problem, and pursuing the more general problem was the better use of the time. GHOST-X addresses GPS-denied state estimation with occlusion tolerance and a target track, on hardware that is easier to iterate on than a throw-launched airframe.

Continuing KESTREL would have meant building an airframe and a flight-test campaign to demonstrate a navigation capability that could be demonstrated more directly, and more rigorously, on a bench. That is a scope decision, and it is worth stating as one rather than letting the project quietly go stale.

06Lessons learned

  • Separate the hard part from its packaging. The interesting engineering was the estimator; the airframe was a delivery mechanism that carried most of the schedule risk.
  • A design target is not a result. Numbers that have not been measured should be labelled as requirements every time they are quoted.
  • Killing a project is a legitimate engineering decision. The failure mode is not stopping, it is continuing out of sunk cost.
  • Treat transients as phases. Designing an explicit throw-detection and recovery mode was more robust than trying to make the nominal controller tolerate the launch transient.
  • Write the closeout. A project without a closeout leaves its lessons undocumented and its status ambiguous.

07What carried forward

KESTREL's central question—how to maintain useful position information without GNSS—helped motivate the narrower, more testable estimator work in GHOST-X. GHOST-X owns the public implementation and test evidence; KESTREL is presented only as the archived design context that helped shape that direction.