Andrii Bidochko

• Updated: August 26, 2026
• 2 min read

Agentic AI for Safety‑critical Multi‑drone Systems: Challenges and Opportunities

Agentic AI for Safety‑critical Multi‑drone Systems: Challenges and Opportunities

Multi‑drone systems are increasingly being deployed for safety‑critical missions such as search‑and‑rescue (SAR) operations and critical infrastructure monitoring. While the potential benefits are substantial, real‑world adoption is limited not only by the performance of autonomous algorithms but also by the difficulty of integrating agentic behavior into professional workflows. Operators must be able to understand, trust, and govern automation under uncertainty, time pressure, and accountability constraints.

This article summarizes the insights from the recent position paper Agentic AI for Safety‑critical Multi‑drone Systems: Challenges and Opportunities (arXiv:2608.21444v1). The paper synthesizes lessons from two ongoing research efforts:

• NAMUR – exploring large‑language‑model‑supported robot control in SAR and firefighting contexts.
• PERSIST – investigating persistent drone operations for monitoring and security at critical infrastructure sites.

Both projects highlight that agentic AI should be treated as a socio‑technical design problem. The design of interfaces, oversight mechanisms, and evaluation practices is as critical as the underlying algorithms.

Key Challenges

1. Human‑centered interaction: Operators need transparent, explainable interfaces that convey the intent and confidence of autonomous agents.
2. Governance and accountability: Clear protocols for intervention, escalation, and post‑mission analysis are essential.
3. Robustness under uncertainty: Agents must handle noisy sensor data, dynamic environments, and time‑critical decision making.
4. Scalability and persistence: Long‑duration missions require reliable hand‑off between agents and continuous situational awareness.

Proposed Human‑Centred Research Approach

The authors advocate an iterative, participatory methodology that includes:

• Stakeholder needs assessment through workshops and field studies.
• Prototype development with incremental agent capabilities.
• Evaluation using realistic scenarios and mixed‑initiative performance metrics.
• Knowledge transfer to other safety‑critical domains.

For more information on our ongoing work, visit the UBOS Tech research page and explore related blog posts such as AI Safety in Autonomous Systems.

Conclusion

Integrating agentic AI into safety‑critical multi‑drone systems requires a holistic approach that balances technical innovation with human factors, governance, and rigorous evaluation. By adopting a human‑centered, participatory design process, researchers and practitioners can develop trustworthy, effective autonomous solutions for high‑stakes missions.

Image: Multi‑drone system operating in a safety‑critical scenario (featured above).

Andrii Bidochko

CTO UBOS

Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.