Error-based and reward-based mechanisms co-occur during real-world motor learning. In motor neuroscience, these two learning mechanisms are often studied in isolation via feedback manipulations in reductionistic laboratory tasks, such as point-to-point reaching movements. We previously demonstrated that in real-world tasks (playing pool billiards), a correction for a large error feedback can be rewarding enough to engage the reward learning mechanism, even in the absence of a reward signal (e.g. ball pocketing). Here, we examine whether these two learning mechanisms could be isolated via sensory feedback manipulations in a gradual perturbation task, while the perturbation-induced errors are smaller. Using mobile brain imaging (EEG) and Embodied Virtual Reality of pool billiards, where visual feedback is at the start veridical with the physical billiard setup, we introduce a gradual visual perturbation of the ball trajectories while keeping the real-world haptics and proprioception unchanged. N=32 participants underwent two sessions, learning gradual visuomotor rotation with either error or reward feedback. Behavioural and neural (Post Movement Beta Rebound over the Motor Cortex) markers recorded throughout the learning process showed a clear within-participant separation between the ways they learned the perturbation in the different sessions with the error and the reward feedback. Our results show how error- and reward-based learning mechanisms could be separated in a real-world task. Our framework and results open the way to AI-controlled manipulation of learning mechanisms to accelerate skill learning in medical (e.g. neurorehabilitation) and vocational uses (surgeons, sports).