Program defects are inevitable during the development and maintenance processes. With the rapid increase of software scales, the number and repair complexity of program defects increase as well, which has caused huge economic loss to enterprises, and becomes the big burden for developers during maintaining. Automatic program repair (APR) techniques have the potential to release developers from heavy debugging tasks, and become a popular research topic recently. This study collected the most recent 94 high-quality publications in this research field. According to analyzing the approaches used for patch generation, APRs are systematically classified into four categories, i.e., search-based, template-based, constraint-based, and statistical-analysis-based APRs. Especially, this study proposed the category of statistical-analysis-based APR for the first time based on the most recent publications, which complements and improves existing taxonomy. Based on existing techniques, the key challenges and insights are summarized for future research. Finally, benchmarks and open-source APR tools are briefly summarized for reference.