R&D / AI drone concept and proof of concept
Companion drone concept with AI shark detection, wearable and iOS alertsDiscovery, system design, AI model training and proof of concept
Proving a follow-me alert drone concept with wristband and phone alerts
Hover Hero, an early-stage startup, had a concept for a compact companion drone for open-water users such as surfers: a drone with a camera and radio module that watches the water for sharks and alerts the user on a phone or wrist-worn device. The founders had a vision, but no architecture, selected hardware or software.
Show that the concept was technically feasible. The requested proof of concept was a drone that hovers and follows a user at about 10 feet, detects moving shapes shown by a video projector on the floor as stand-ins for real threats, sends haptic alerts to several wristbands and shares snapshots with an iOS app that also adjusts altitude and basic controls.
- Product
- Companion drone concept with AI shark detection, wearable and iOS alerts
- Buyer
- Early-stage founders who need to show a drone or wearable concept is technically feasible
- Engagement
- Discovery, system design, AI model training and proof of concept
- Indeema scope
- Discovery workshops, Requirements and system design, Proof-of-concept drone, Follow-me behavior, AI detection and alert flow, Development plan
The product need
What had to become possible
Show that the concept was technically feasible. The requested proof of concept was a drone that hovers and follows a user at about 10 feet, detects moving shapes shown by a video projector on the floor as stand-ins for real threats, sends haptic alerts to several wristbands and shares snapshots with an iOS app that also adjusts altitude and basic controls.
Ownership, made clear
Where the work started and what Indeema owned
Starting system
An idea, not a system
At the start there was only a high-level description of how the drone should behave and interact with users; architecture, components, firmware and software all had to be defined.
Indeema's drone hardware platform
The proof of concept was built on a hardware platform Indeema already had, rather than new custom electronics.
Open-source flight software
Flight control used open-source flight software, Betaflight or iNAV, instead of custom firmware at this stage.
Indeema owned
Discovery workshops
Run workshops with the founders to clarify the product vision, prioritize features and document new use cases that surfaced along the way.
Requirements and system design
Write a functional requirements document and a system design document covering architecture, hardware and firmware stack, connectivity logic, edge-processing design and the interaction with an iPhone app and Apple Watch.
Proof-of-concept drone
Integrate open-source flight software on the existing platform and test stability, flight control and basic communication.
Follow-me behavior
Run real-world tests with the client to choose the hover and follow distance and settle on a follow-me mode that fits the hardware.
AI detection and alert flow
Train a neural network in Indeema's lab to detect sharks in drone-view imagery, and define the recognition flow and notification logic from detection on the drone to wearable and phone alerts.
Development plan
Produce a plan for further development, including future feature extensions.
Worked with
Flight platform
Existing drone hardware with flight-control and connectivity modules; project silicon includes the STM32F405, Silicon Labs EFM8BB21 and Espressif ESP32-C6.
Wearables
Haptic alerts to multiple wristbands, with Apple Watch interaction defined in the system design.
iOS app
Snapshots from the drone, altitude adjustment and basic controls on the phone.
AI detection model
A YOLO11n object-detection network trained on annotated drone-view images, compact enough for processing on the drone.
Product constraints
What made the engineering difficult
- 01
Choosing the hover and follow distance
The distance at which the drone hovers or follows had to balance smooth movement against hardware and budget limits.
- 02
Sharks seen from the air
From above, a shark can be a faint shape under glare, waves and reflections, so the model needed training data from a drone's point of view.
- 03
Detection while moving
Recognizing objects from a moving drone brings camera shake, motion blur and changing lighting, all of which affect on-board detection.
- 04
A vision without a specification
The concept had no architecture or component choices, so feasibility work had to start by defining the system itself.
- 05
Keeping the proof small
The proof of concept had to show the core interaction without the cost of custom electronics and firmware.
Engineering judgement
Decisions that shaped the product
Reuse an existing hardware platform
Build on Indeema's existing drone hardware to speed up early validation.
Defer custom firmware
Use Betaflight or iNAV for flight control in the proof of concept, leaving room for custom firmware later.
Simulate the threat
Use a projector showing moving shapes as targets, so detection and alerts could be tested in a controlled setting before any open-water work.
Train the model in our own lab
Annotate drone-view imagery and train and evaluate the shark-detection network in Indeema's lab, measuring precision and recall at several input sizes.
Keep detection on the drone and lightweight
Choose a compact neural network, YOLO11n, suited to on-board processing so the alert path stays responsive.
Decide distance with field tests
Settle the follow-me configuration through real-world tests with the client rather than on paper.
Delivered product capability
What the product can do
- 01
System architecture
A structured architecture with the hardware and firmware stack defined.
- 02
Engineering documentation
Functional requirements and system design documents for the concept.
- 03
Selected drone platform
A drone platform with flight-control and connectivity modules chosen for the proof of concept.
- 04
Follow-me mode
A follow-me configuration chosen through tests with the client that stays within the platform's resources.
- 05
Trained shark-detection model
A neural network trained and evaluated in Indeema's lab to detect sharks in drone-view images, ahead of integration on the drone's hardware.
- 06
Alert interaction design
Defined mobile and wearable interaction, recognition flow and notification logic for the simulated scenario.
- 07
Development plan
A plan for the next development stages and future features.
Product change
What changed for the product team
Hover Hero moved from a high-level idea of a companion drone to a documented architecture, a selected platform, a follow-me configuration tested with the founders, a shark-detection neural network trained in Indeema's lab, a defined alert flow and a development plan. This is a proof of concept: the alert path was exercised with simulated targets projected onto the floor, and the model was evaluated on images rather than on sharks in open water.
For a founder with a drone or wearable concept, the case shows how a short R&D phase can separate what has to be proven now (flight behavior, the alert path, a first trained AI model, detection in a controlled scenario) from what can wait (custom firmware and field-ready perception), using existing hardware and open-source flight software.
- The founders received a documented system architecture, engineering documents and a development plan for the companion-drone concept.
- A follow-me configuration was chosen through real-world tests with the client on an existing drone platform.
- A shark-detection neural network was trained and evaluated on drone-view images in Indeema's lab.
Product evidence
A closer look at the product
Technology, in context
Swift · C/C++ · Python · Betaflight / iNAV · STM32F405 · Silicon Labs EFM8BB21 · ESP32-C6 · YOLO11n · Computer vision / neural network training
A comparable product problem?
Hover Hero: companion drone with AI shark detection
Bring the product context, the system that already exists and the engineering decision you need to make.