A swarm of non-holonomic Dubins-car like robots has to autonomously organize itself into a densest net over a cross-section of a corridor environment. Then the swarm has to keep this formation, while moving along the corridor with a pre-specified speed. Any robot measures its own speed and, in its local frame, has access to the relative locations of the objects within a finite sensing range. The robots do not differentiate the peers, are unaware of the swarm's size and corridor's width, do not carry communication facilities and cannot play distinct roles in the team. A computationally inexpensive, decentralized and distributed control strategy is presented that solves the mission. The performance of the proposed algorithm is justified via a mathematically rigorous global convergence result for the corridors with straight parallel walls and is demonstrated via computer simulation tests for a variety of more intricate cases.