A new decentralized control method, named Regular Tetrahedron Formation (RTF) strategy, is presented for a swarm of simple robots operating in three dimensional space. This strategy is based on virtual spring mechanism and basically, allows four neighboring robots to autonomously form a Regular Tetrahedron (RT) regardless of their initial positions. RTF strategy is made scalable and applied for various sizes of swarms through a dynamic neighbor selection procedure. Thus, each robot's behavior in each time step is only dependent on the local position information of three dynamically selected neighbors. In addition, an obstacle avoidance model suitable for swarm maneuvering is also introduced. Algorithm is studied with computational experiments which demonstrated that it is effective and practical.