Task engagement is a key factor in sustaining patients' participation in rehabilitation. Adaptive task difficulty level adjustment techniques are designed to determine the appropriate exercise difficulty level in where subjects are appropriately challenged and engaged without causing any distress. Such adaptive difficulty adjustment within rehabilitation tasks has the potential to individualise training. In this study, the authors have compared two dynamic difficulty level adjustment algorithms, partially ordered set master (POSM) and increment/decrement one level (IDOL), those change the difficulty levels for each individual adaptively based on his/her performance. These two algorithms are integrated into the robot-assisted rehabilitation system, RehabRoby, and their functionality is explored via a small user study with 20 healthy subjects. The subjects were asked to perform a computer-based fruit picker game using RehabRoby under these two algorithms. The impacts of the adaptation algorithms are evaluated in terms of engagement of the subjects by looking at their physiological signals, performance (score), and survey results. Experiments show that although POSM on the average suggests easier difficulty levels to the subjects than IDOL, the subjects experience a wider range of difficulty levels in POSM that may help them to become more engaged in the game, which can also be observed by lower skin temperatures of the subjects.