Short updates on my latest research.

August 14, 2026 · 3 min read

Swarm-Steward: Talking to Drone Swarms in Plain Language

Our Swarm-Steward system lets a single non-expert operator coordinate fleets of drones and ground robots using natural language — no joysticks or programming required. The key idea is a hierarchical LLM-based multi-agent design that separates planning from actuation: specialized agents ground each request in real map features and telemetry, while a deterministic safety gate (geofencing, altitude and separation limits) vets every command before execution, with optional operator preview. To keep reasoning scalable, we use dual retrieval-augmented generation over map features and telemetry state — effective even with 10,000 map features. In end-to-end tests with swarms from 5 to 500 drones, LLM token cost stayed nearly constant, with 92.9% task success across 280 prompts.

2026 · IEEE ICUAS IEEE PDF

A. Jarabo-Peñas, J. Bravo-Arrabal, E.G.A. Rolland, A.L. Christensen

August 14, 2026 · 2 min read

Voice-Commanded Robots for Real Search & Rescue Logistics

Search and rescue teams can't afford to become robot programmers mid-mission — so in this work we took our voice-driven control out of the lab and into real-world conditions. We deployed an LLM-enabled, voice-commanded UGV for logistics support in SAR missions: operators with no robotics expertise simply speak natural-language commands, and the robot translates them into safe, context-aware navigation grounded in real-time mission data. The paper reports the deployment architecture, the challenges of operating in the field, and lessons for making autonomous ground robots practical tools for first responders.

2025 · IEEE SSRR IEEE PDF

A. Jarabo-Peñas, J. Bravo-Arrabal, D. Lin-Yang, F. Pastor, R. Ladig, J.J. Fernández-Lozano, A.L. Christensen, A. García-Cerezo

← Back to Publications