auTonomous Robotic aId For increasing FIrst responDers efficiency

Project Information

TRIFFID

Grant agreement ID: 101168042

DOI

10.3030/101168042

EC signature date 16 July 2024

Start date 1 October 2024

End date 30 September 2027

Funded under

• Civil Security for Society

Total cost

€ 4 098 690,00

EU contribution

€ 4 098 690,00

Coordinated by

CHAROKOPEIO PANEPISTIMIO
Greece

Periodic Reporting for period 1 - TRIFFID (auTonomous Robotic aId For increasing FIrst responDers efficiency)

Reporting period: 2024-10-01 to 2026-03-31

Summary of the context and overall objectives of the project

First responders increasingly face large-scale wildfires, floods and earthquakes whose frequency and severity are rising with climate change and urbanisation. Operating in these environments is dangerous, time-critical and information-poor: responders must make decisions quickly while exposed to fire, unstable structures, toxic atmospheres and rising water. TRIFFID addresses this challenge by giving emergency teams a trustworthy, AI-powered robotic assistant that can be sent into hazardous areas in their place.

TRIFFID develops an integrated system combining a legged unmanned ground vehicle (UGV) and an unmanned aerial vehicle (UAV) that cooperate to explore disaster sites, build situational awareness and relay it to a command station through an intuitive interface. The project pursues four objectives: (O1) increase operational efficiency through an autonomous, AI-powered hybrid legged UGV+UAV platform for remote reconnaissance; (O2) enhance situational awareness without endangering people, through robotic perception, Earth-observation data analysis and human-robot interaction; (O3) foster the responsible adoption of robotic/AI systems in field operations and recommend appropriate standards and policies, ensuring human safety and compliance with ethical and legal guidelines; and (O4) build the complete TRIFFID system and demonstrate it in real-world fire, flood and earthquake scenarios.

Social sciences and humanities are central to the project: human-factors research, participatory design with first responders, and societal and ethical impact assessment shape the system so that it is not only technically capable but genuinely usable and accepted by the people who will rely on it. The pathway to impact runs from these foundations, through system integration, to three field pilots with civil-protection agencies in Europe, where the benefits for responder safety and effectiveness will be validated.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

During the first 18 months TRIFFID built the technical foundations of the system across mission planning, perception, human-robot interaction, human factors and system integration.

For autonomy and control (O1), the consortium developed a hierarchical mission-planning system based on a timeline-based planning and execution framework (PLATINUm) coupled with the ROXANNE reactive acting system, enabling dynamic task decomposition and real-time plan adaptation for both platforms. Autonomous navigation was implemented for the DJI Matrice 30T UAV (waypoint missions with GPS guidance and obstacle avoidance) and the Unitree B2-W legged UGV (SLAM-based navigation using LiDAR and RGB-D sensing with 3D mapping). An integrated safety-monitoring framework runs in parallel with planning, covering fault detection and a three-tier recovery strategy (dynamic re-planning, safe-stop-in-place and supervised return-to-base), alongside a resilient communications backbone combining Wi-Fi, 4G/LTE and Starlink fallback. These results are consolidated in deliverable D3.1.

For perception and situational awareness (O2), the team created the Incidents1M-Seg benchmark dataset — 170,272 annotated instances across 15,033 disaster images and 63 classes — and trained models reaching state-of-the-art accuracy: 95.58% mIoU for building-damage assessment on RescueNet, 95.80% mAP50 for aerial object detection and a gesture-recognition system exceeding 90% correct classification. Multi-modal sensor fusion extends perception beyond the visible spectrum, and an AR-based command-and-control ground station presents three complementary views (2D operational map, 3D LiDAR point cloud and AR-augmented camera feeds). These results are reported in D4.1.

For responsible integration (O3), a human-factors analysis using the EAST (Event Analysis of Systemic Teamwork) methodology characterised the task, social, information and knowledge networks of disaster-response teams, and a participatory-design toolkit was produced (D5.1). End-user requirements were systematically elicited and prioritised (D2.1) and realistic pilot scenarios for the three use cases were defined (D2.3).

For system building (O4), end-user needs were translated into a complete system specification and architecture (D6.1) comprising 43 functional and 8 non-functional requirements, and the hardware and middleware integration baseline was established (D6.2) including the selected UAV and UGV platforms, a combined RGB/thermal/LiDAR/gas/acoustic sensing suite and a dual-computing architecture (Intel NUC and NVIDIA Jetson AGX Orin). All reporting-period milestones were achieved on schedule, with individual modules validated in the laboratory ahead of system-level integration and the field pilots.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

TRIFFID has produced several results that advance the state of the art in robotics and AI for disaster response.

The Incidents1M-Seg dataset is, to the consortium’s knowledge, the largest pixel-level annotated benchmark of real disaster imagery to date, providing a new resource for training and evaluating perception models in emergency contexts. Building on it, the project’s Earth-observation damage-assessment and detection models achieve accuracy above current published baselines (95.58% mIoU on RescueNet; 95.80% mAP50 aerial detection), demonstrating that reliable automated scene understanding is feasible in cluttered, degraded disaster environments.

The combination of timeline-based deliberative planning with a reactive acting layer advances autonomous mission management for heterogeneous UGV+UAV teams operating under uncertainty and intermittent communications, an area where existing systems typically address ground or air platforms in isolation. The safety-monitoring and fault-recovery framework contributes a structured approach to dependable autonomous operation in unstructured settings.

Beyond individual components, TRIFFID’s integration of human-factors evidence and participatory design directly into system specification represents a methodological advance over technology-led approaches, increasing the likelihood of real operational uptake. To ensure further uptake and success, the project has identified a portfolio of 14 exploitable results with individual exploitation plans and business cases, six Key Exploitable Results, and the principal needs going forward: completion of the field pilots, demonstrable operational relevance for civil-protection users, supportive regulatory and standardisation frameworks, and sustained industry partnerships to bridge the gap from prototype to market in a sector with limited current maturity.