Machine learning powered robotic swarms promise scalable, adaptive solutions across logistics, healthcare, public services, and manufacturing. Yet as autonomous agents increase in number and complexity, new risks emerge. Even when individual robots operate correctly, their interactions can generate unexpected system level behaviors that are difficult to predict, verify, or certify. This scholarly research presentation introduces SAFESwarm, a structured assurance framework designed to address these collective risks.
Mayank Jha of Amazon Robotics presents an end to end lifecycle driven methodology tailored specifically for learning enabled multi agent systems. SAFESwarm extends existing assurance models for autonomous systems by focusing not only on single agent correctness but on emergent swarm behavior. The framework organizes assurance activities into six stages: safety scoping, requirement elicitation, data governance, model development, probabilistic verification, and deployment monitoring.
Each stage produces traceable artifacts that contribute to a defensible safety case. Rather than relying solely on post deployment observation, SAFESwarm enables formal specification and validation of swarm level properties such as collision avoidance, graceful degradation, and human aware coordination. This structured approach supports verifiable and certifiable deployment of collaborative robotic collectives in complex real world environments.
The session demonstrates the framework through a realistic case study involving a swarm of assistive robots operating a public cloakroom in a human populated space. Using simulation based testing, statistical model checking, and fault injection analysis, the research shows how probabilistic safety guarantees can be established and maintained even under communication disruptions and hardware faults. Results indicate a significant reduction in unintended agent interactions while preserving operational efficiency.
From an information systems and operations management perspective, SAFESwarm provides a practical foundation for deploying trustworthy autonomous systems at scale. As organizations increasingly rely on distributed intelligent agents, structured assurance becomes essential not only for safety compliance but also for stakeholder trust and long term system resilience.
Author and Affiliation
Mayank Jha, Amazon Robotics
Delivered virtually within the Information Systems and Operations Management track at the SAM International Business Conference, this session offers valuable insight for researchers, engineers, and management professionals working in robotics, autonomous systems, and trustworthy artificial intelligence. If you are exploring how to manage risk, verification, and certification in learning enabled multi agent environments, this research provides a structured and actionable roadmap. Learn more about this presentation and register to attend at www.samnational.org/conference.