We have just concluded an extraordinary week of technology testing at Ol Pejeta Conservancy and Mpala Research Centre in Laikipia, Kenya — and what a ride! As part of the WildDrone M36 Field Trip Workshop (EU Horizon Europe MSCA Doctoral Network), our international team of researchers, wildlife rangers, and engineers pushed the frontiers of conservation technology: flying solar-powered glider drones, deploying autonomous multi-drone swarms, and applying edge AI to massive amounts of ecological field data.
The global interest was enormous: we were honoured to host television film crews from Sweden, Germany, and Kenya documenting how autonomous robotics and embodied AI are transforming wildlife protection on the frontlines.
🦏 The Conservation Challenge: Finding Rhinos in Dense Bush
Ol Pejeta Conservancy is home to Africa’s largest population of black rhinos and the world’s last two remaining northern white rhinos. Yet monitoring these magnificent animals in dense bushlands, thick acacia woodlands, and along riverine corridors remains one of conservation’s toughest operational hurdles. When an animal is partially occluded by tree canopies or camouflaged against cluttered brush, detection from a single overhead vantage point frequently fails.
Traditional manual drone operations require constant joystick piloting, struggle with single-angle blind spots, and impose heavy cognitive workload on field rangers. As we highlighted during our presentation at the Global Conservation Tech & Drone Forum in Nairobi: “How to find rhinos with drones? Not that easy...”
🗣️ WildVoice RhinoScout: Swarm-Steward in the Savannah
To overcome the single-viewpoint limitation, we deployed WildVoice RhinoScout, built directly on our Swarm-Steward natural-language multi-robot coordination architecture (accepted at IEEE ICUAS 2026). Swarm-Steward combines hierarchical LLM-based multi-agent reasoning with spatial grounding, telemetry-aware context retrieval, and deterministic safety validation.
Integrated with the WildBridge robotics middleware, WildVoice RhinoScout allows a human operator (such as an ecologist or wildlife ranger) to command an entire autonomous drone swarm through intuitive spoken language:
“Scan between the trees across the northern riverbank.”
“Focus near the bushes and report any large mammal movement.”
The system interprets the high-level intent, grounds the references against real-time GPS telemetry and spatial map features, and autonomously translates the command into coordinated search paths across up to 5 DJI Mini 4 Pro drones flying simultaneously.
🎯 Multi-Angle Vision & On-Board YOLO Detection
Each drone in the swarm runs a lightweight YOLO-based detection pipeline trained specifically for rhino detection under heavy occlusion and cluttered vegetation. By dynamically adapting their relative flight formations and viewing angles:
- Nadir & oblique perspectives: Simultaneous top-down and angled viewpoints allow the system to confirm detections that are completely hidden from a single angle.
- 3D Biometrics & Health Assessment: Synchronized multi-perspective imagery supports downstream photogrammetric body-condition scoring (age, sex, nutritional state, pregnancy status).
- Low Disturbance: Autonomous trajectory optimization ensures quiet, non-invasive observation without stressing sensitive wildlife.
⚡ Field Results & Lessons Learned
Field deployments in Laikipia proved the robustness of the voice-driven swarm architecture:
- Reliable Spoken Coordination: Voice interaction and multi-agent intent translation executed consistently throughout the trials, enabling rapid re-routing of swarm members on the fly.
- Offline Detection Accuracy: The YOLO detection model demonstrated exceptional performance on recorded field datasets across open grassland, shrublands, and dense riverbanks.
- Overcoming Video Bottlenecks: While early missions contended with intermittent live video bandwidth over savanna terrain, we resolved the transmission pipeline in the final days, stabilizing real-time 1080p video streaming at 10 fps.
The curated rhino detection dataset along with our trained YOLO weights will be released open-source to support the global conservation robotics community, and a full research paper detailing the system architecture and field benchmarks is currently in preparation.
🛸 Solar Gliders, Tethered Balloons & Multi-Perspective Swarms
WildVoice RhinoScout was part of a much broader, highly collaborative research effort across the WildDrone consortium during the 9-day field trip. A wide spectrum of cutting-edge conservation robotics projects was tested across Ol Pejeta and Mpala Conservancies:
- Solar-Powered Gliders (Helios & MACE v0): Teams from SDU and the University of Bristol field-tested lightweight, solar-powered fixed-wing gliders designed for ultra-long-endurance Beyond Visual Line of Sight (BVLOS) monitoring. The tropical sun in Laikipia provided an ideal proving ground to assess solar energy harvesting, flight dynamics, and endurance under realistic conservation conditions.
- WildSwarm & Multi-Perspective Herd Tracking: Edouard Rolland (SDU) deployed autonomous swarms of DJI Mini 4 Pro drones to simultaneously capture synchronized top-down (nadir) and horizontal (oblique) perspectives of plain zebras and giraffes. This multi-viewpoint telemetry enables 3D animal reconstruction, posture estimation, and social interaction mapping without human disturbance.
- WildDisturbance & Swarm Acoustics: Researchers measured the acoustic footprint of multi-drone swarms across different flight altitudes and formation spacings in open savannah environments, recording wildlife behavioural responses to establish empirical guidelines for noise-minimizing drone operations (submitted to Quiet Drones 2026).
- Tethered Balloon Aerostats (WildBalloons): Helium-filled tethered balloon platforms equipped with high-resolution payloads were evaluated for zero-rotor-noise, persistent, hours-long surveillance over watering holes and migration corridors.
- Aerial Photogrammetry on 86 Rhinos (WildRhino): Camille Rondeau Saint-Jean (SDU), working alongside Ol Pejeta’s expert rangers, collected aerial video and laser-rangefinder photogrammetry on 86 individual black and white rhinos (including direct physical measurements during ranger-led translocations) to calibrate non-invasive biometric models for body condition, age, and pregnancy.
- Adaptive Search vs. Lawnmower Routing (WildSearch): University of Bristol and Southampton researchers benchmarked entropy-driven search algorithms balancing exploration and re-viewing against traditional lawnmower coverage over wildlife watering holes.
🏛️ Nairobi Outreach & United Nations Visit
Following the fieldwork in Laikipia, our team travelled to Nairobi to present our findings at the Global Conservation Tech & Drone Forum (GCTDF) and the Women in Conservation Forum. A central pillar of the WildDrone mission is capacity building and knowledge transfer:
- We conducted structured robotics and computer vision demonstrations with students at Kenyatta University (KU), opening pathways for the next generation of African conservation technologists.
- Team members visited the United Nations Complex in Nairobi (UNON) to discuss how autonomous aerial systems and embodied AI can inform international biodiversity monitoring and environmental policy.
🏖️ Unwinding on the Kenyan Coast: Watamu & the Indian Ocean
After an intense, sun-drenched fortnight of field experiments, data crunching, and forum presentations, we wrapped up the expedition by spending our final week on a well-deserved vacation along the stunning tropical coast of Kenya.
From relaxing on the white sand beaches of Watamu and exploring the turquoise waters of the Indian Ocean to navigating coastal mangrove channels and soaking in the local Swahili culture and carved architecture, it was the perfect adventure to recharge before returning to Europe with hard drives full of data!
March 2026 · WildDrone Field Trip WildDrone EU Swarm-Steward (ICUAS 2026) Ol Pejeta
WildDrone is an MSCA Doctoral Network funded by the European Union’s Horizon Europe research and innovation programme under grant agreement no. 101071224.