Mechanical Engineering Department

National University of Singapore

Research Interest: small unmanned aircraft design, applied aerodynamics, aerial robotics, unmanned aerial systems and its applications, especially: vision-based navigation, perception and sensing for UAS, counter-UAS, multi-agent and swarm of UAVs.

Speech Title:Topology-Safe Air-to-Air Interception via “Go”-Inspired Metrics and Multi-Agent Reinforcement Learning

Abstract:The proliferation of agile, low-cost UAVs turns airspace defense into a cooperative–adversarial multi-agent problem where multiple defenders must coordinate against fast, evasive intruders under sensing noise and delays. While circular-formation controllers can achieve geometric closure, they do not certify when an intruder is truly “trapped”— i.e., when no dynamically feasible escape corridor remains. This manuscript addresses that gap with a “Go”-inspired, two-layer framework that unifies topological safety certification with control and learning. At the core, “Go” notions of liberties and eyes are mapped to continuous-space escape corridors and residual cut capacity, yielding three online metrics—largest angular gap, escape-corridor count, and normalized residual cut capacity—that decide entrapment relative to intruder dynamics. The encirclement model introduces a direction relevance gating mechanism that enforces the Go principle that a single stone cannot surround alone, naturally requiring multi-agent cooperation. A MAPPO-based learning layer, trained under centralized-training/decentralized-execution with curriculum-scheduled capture thresholds, learns cooperative interception policies that progressively tighten the encirclement ring. Simulation results on a 2D air-combat environment demonstrate that a team of three interceptors achieves a 63% capture rate against an intelligent evader under strict, multi-directional blockade criteria, with average evader liberty reduced from 0.76 to 0.25 and blocked directions increasing from 2.25 to 6.04 out of 8 over 5 million training steps.

Bio:

Dr. Sutthiphong Srigrarom, also known as “Dr.Spot”, is at Mechanical Engineering Department at National University of Singapore. He received his PhD from University of Washington, USA in 2002. Dr.Spot is also a visiting associate professor at Institute of Flight System Dynamics, Technical University of Munich in Germany. Dr.Spot’s research work is mainly on small unmanned aircraft design, applied aerodynamics, aerial robotics, unmanned aerial systems and its applications, especially: vision-based navigation, perception and sensing for UAS, counter-UAS, multi-agent and swarm of UAVs