Simulated Teacher - progress visualization

The main idea is to emulate the interactions of a robot with the surrounding environment over a long period in a single context scenario (office, kitchen, etc.), where a robot would be expected to learn continually and adaptively. In such a human-in-the-loop open-ended learning scenario, the interaction between the human user and the robot unfolds through a sequence of actions: teach, ask, and correct. The teaching process commences with the human user introducing an instance of a new object category. The provided instance serves as the representative of the category, providing the robot with an initial understanding of that category. After the learning phase, the robot is tested on recognizing new instances of the categories it has learned so far. This critical step evaluates whether the robot is susceptible to catastrophic forgetting, a phenomenon wherein learning new information interferes with the previously learned categories. If the accuracy surpasses a predefined threshold, $\tau$, indicating a satisfactory level of retention and adaptation, the system proceeds to the next step. At this point, the human user introduces a new object category, and the teaching process repeats. The agent continuously learns and adapts to the expanding set of categories introduced by the user.