Physical environment perception, modification, and measurement are challenging in constrained space at a small scale due to the lack of accessible sensors and related equipment. Here, we report a novel environmental learning strategy that utilizes microrobot swarms' responses to physical environments and interactions to detect useful details about surroundings, such as physical sizes and environmental viscosity. Inspired by natural collective behaviors, magnetic microparticle-based microrobots are organized into multiple cilia-like swarms. Under wireless magnetic actuation, these swarms demonstrate attractive performances in crowded and constrained spaces. By observing their responses to different surroundings, such as passing through different obstacles, it is feasible to obtain the detailed parameters of the obstacles and learn information on the surface, including slope angles and heights. Swarm pattern generation is also a great reflection of environmental viscosity. Detailed physical analysis and theoretical explanation are conducted to demonstrate the effectiveness. Several experiments are conducted to verify the proposed strategy. This work paves a valuable path to learning and modifying the environment and surroundings at a small scale.