Understanding the mechanisms underlying the diversity of tumor invasion dynamics, including single-cell migration, multicellular streaming, and the emergence of various collective migration patterns, is a long-standing problem in cancer research. Here we have designed and fabricated a series of microchips containing high-throughput microscale tracks using protein repelling coating technology, which were then covered with a thin Matrigel layer. By varying the geometrical confinement (track width) and microenvironment factors (Matrigel concentration), we have reproduced a diversity of collective migration patterns in the chips, which were also observed in vivo. We have further classified the collective patterns and quantified the emergence probability of each class of patterns as a function of microtrack width and Matrigel concentration to devise a quantitive "collective pattern diagram." To elucidate the mechanisms behind the emergence of various collective patterns, we employed cellular automaton simulations, incorporating the effects of both direct cell-cell interactions and microenvironment factors (e.g., chemical gradient and extracellular matrix degradation). Our simulations suggest that tumor cell phenotype heterogeneity, and the associated dynamic selection of a favorable phenotype via cell-microenivronment interactions, are key to the emergence of the observed collective patterns in vitro.