Post-Award Roadmap
The 48-hour PoC artifact demonstrates the AiTR algorithmic core, the
robustness story (perturbation sim), and the integration architecture
(stubbed CROWS / thermal / TAK adapters with clean ABCs). The post-award
plan turns the demonstrator into a fielded prototype suitable for the
live-fire range demonstration.
This document is the honest "what we don't have yet" companion to the
proposal. It is intended to be read by DIU evaluators in conjunction with
docs/architecture.md and docs/integration-contract.md.
Honest baseline (today)
- Team: 1 PI (Argus Defense) + Claude Code as development partner.
- Hardware in hand: Apple Silicon developer machine. No CROWS, no
supplementary sensors, no edge compute target.
- Data: Open-source / CC-licensed UAS clips only. No Argus-licensed
UAS imagery yet.
- Model: YOLOv8n on COCO weights with UAS-proxy class filtering. No
UAS-specific fine-tuning.
- Integration: Software-only PoC. CROWS / thermal / TAK / GVA are
documented stubs.
- In-person pitch and live demonstration on Argus laptop + projector.
- Q&A on integration plan, IP, ethics, and partnership pipeline.
- Schedule alignment with DIU on the 3-month post-award window.
Post-award 3-month plan
| Workstream |
Deliverable |
| Hiring |
1 ML engineer (UAS detection fine-tuning); 1 systems engineer (CROWS / integration). |
| CROWS access |
Engage Kongsberg US (CROWS-J/M153 OEM) for vendor surrogate + interface specs. Parallel: government FMS / NSWC channels. |
| Supplementary sensors |
Source EO telephoto (Edmund / MOOG class) + cooled MWIR (FLIR Boson+ / Tau 2 class) for 600 m detection envelope. SWaP-fit against M153 base. |
| Compute target |
NVIDIA Jetson AGX Orin (or equivalent). Port pipeline; convert YOLO to TensorRT. |
| Pipeline |
Wire the existing RTSPSource and ThermalSource stubs. EO+IR detector fusion strategy. |
| Stabilizer |
Promote FeatureStabilizer with low-pass trajectory smoothing + IMU fusion. |
Month 2 — UAS fine-tuning + tracker hardening + ranging upgrade
| Workstream |
Deliverable |
| Dataset |
Argus-licensed UAS imagery acquisition (commercial sources + government-shared data sets). |
| Model |
Fine-tune RT-DETR (Apache-2.0) or YOLOv9 (replaces Ultralytics-AGPL for production licensing). Hard-negative mining against cluttered backgrounds and bird confusion. |
| Tracker |
Swap IoUTracker for ByteTrack/BoT-SORT for dense scenes. Retain IoUTracker as the failsafe path. |
| Ranging |
Add platform-motion triangulation (turret slew parallax). Optional limited-duration laser ranging (CSO permits). |
| Range bench |
Indoor range testing 50–200 m initially; outdoor 200–600 m by month-end. |
Month 3 — live-fire integration + range demonstration
| Workstream |
Deliverable |
| Live-fire range |
Government range slot scheduled. Argus delivers system + on-site engineer + remote support. |
| HITL workflow |
Operator handbook + checklist + dry runs with USMC / Army operators if available. |
| Audit + reporting |
Verification & validation plan; written safety case; post-engagement audit log review. |
| Demonstration |
Live UAS engagement scenario: Group 1 quadcopter at 100 m, multi-target option, adverse conditions injected. |
| Documentation |
Final technical report; transition recommendation; IP & licensing assignment. |
Partnership pipeline
| Partner |
Role |
| Kongsberg US (M153 OEM) |
CROWS surrogate access; interface specs; integration review |
| NSWC / Army C5ISR Center |
Government-side range access; SME consult; FMS channel |
| FLIR / Teledyne |
Cooled MWIR sensor sourcing |
| Edmund / MOOG |
EO telephoto + gimbal integration |
| NVIDIA |
Jetson AGX Orin + DGSC partnership |
| Anduril / Skydio / Shield AI |
Future C-UAS partner discussions (defense-tech ecosystem) |
What this is not
- This is not a commitment that all partnerships will close in three
months. The roadmap shows the credible path and the decision points.
- This is not a guarantee of live-fire success on every shot — the
CSO's stated engagement range (≥100 m) is achievable; performance
beyond 200 m for Group 1 lateral movement (Part 2) is post-Part-1 work.
- This is not a hand-wave on safety. Live-fire integration is gated on
written safety case and government range acceptance.
Beyond Month 3
- Part 2 alignment: Multi-target engagement at lateral 7 m/s, ±10° to
90° elevation, 360° azimuth. Part 2 work plan is a separate scope.
- Part 4 alignment: TAK + post-quantum encryption layer over the
telemetry stream. Already stubbed in
integrate/telemetry_sink.py.
- Production hardening: ATO-style V&V, environmental qualification,
long-tail data collection, operator certification program.