AbstractIn collaborative haptic training systems, a novice operator is interfaced with an expert operator and cooperatively performs some task on a real/virtual environment. Most control architectures for collaborative haptic training systems do not consider the switching task dominance together with investigating overall stability in the presence of nonlinear dynamics and uncertainty. In this paper, a theoretical framework is presented for switching task dominance in collaborative haptic training systems based on supervision and intervention of the expert operator. To that effect, the novice operator performs the operation with as little as possible interference haptic signals in the normal operational conditions. On the other hand, the expert operator is able to intervene the operation to guide the novice operator when it is necessary. The most challenging part of controller design for such systems is to provide the mentioned supervisory framework in a way that the stability of interaction between the operators and the system is ensured with acceptable task performance in various operational conditions. This work offers a variable‐gain dual robust control scheme to address the above problem. The key idea is that the tracking gain of each controller is adjusted in real‐time to switch the task authorities. It is verified that the input‐to‐state stability property is satisfied for each subsystem. Then, the overall stability is proved by leveraging the small gain theorem. Finally, the functionality and performance of the suggested control architecture is demonstrated through simulation and experimental studies.