偽の仲間に探索を妨げられるドローン群、実体確認と未確認IDの隔離で防御を検証

Sybil-Resilient Multi-Drone Exploration of Indoor and Confined Spaces

When a team of drones explores a building together, each drone tells the others which frontier it is going to and shares the map it has built. This avoids duplicated effort, but it also means that the team trusts every identity from which it receives messages. In a Sybil attack, one compromised drone creates fake identities, and these fake teammates claim frontiers and send fabricated scans, so the real drones leave parts of the building unexplored and end up with walls in their maps that do not exist. We study five such attackers, including one that seals off the openings between floors, and five levels of defence that need no security infrastructure, only the drones' own sensors and maps: an audit of received map data against what a drone sees itself, a check that a body is present where an identity claims to be, trust only for identities whose body has been seen, a quarantine for identities that no drone has observed, and a rule that binds each observed body to one identity. Tests in simulation cover 70 maps of seven kinds, from offices and tunnels to three-storey buildings and caves. Without a defence, one compromised drone leaves on average up to 25% of a map unexplored (45% in offices), and the floor-sealing attacker cuts the explored part of a three-storey building to 82.3%, or 34.6% on the worst map. The quarantine brings exploration back to the level of an unattacked team in every setting, and in most settings it also brings the false data back to that level. The exceptions are two colluding attackers, fakes placed next to real drones, three of the ten caves and a team of 16 drones, and binding bodies to identities fixes the last three. Over 2,350 missions, an honest drone was revoked by a teammate only twice.