Programmable robots are engaging and fun to play with, interact with the real world, and are therefore well suited to introduce young learners to programming. Introductory robot programming languages often extend existing block-based languages such as Scratch. While teaching programming with such languages is well established, the interaction with the real world in robot programs leads to specific challenges, for which learners and educators may require assistance and feedback. A practical approach to provide this feedback is by identifying and pointing out patterns in the code that are indicative of good or bad solutions. While such patterns have been defined for regular block-based programs, robot-specific programming aspects have not been considered so far. The aim of this paper is therefore to identify patterns specific to robot programming for the Scratch-based mBlock programming language, which is used for the popular mBot and Codey Rocky robots. We identify: (1) 26  bug patterns, which indicate erroneous code; (2) three code smells, which indicate code that may work but is written in a confusing or difficult to understand way; and (3) 18 code perfumes, which indicate aspects of code that are likely good. We extend the LitterBox analysis framework to automatically identify these patterns in mBlock programs. Evaluated on a dataset of 3,540 mBlock programs, we find a total of 6,129 instances of bug patterns, 592 code smells and 14,495 code perfumes. This demonstrates the potential of our approach to provide feedback and assistance to learners and educators alike for their mBlock robot programs.