Enhancing the vision performance of a demonstration-based robot programming framework through continuous learning and multi-modal data fusion

Continuous learning is an approach to machine learning where an algorithm is trained to perform different tasks over time. It acquires knowledge and skills from different past experiences to improve its performance in the future. To achieve continuous learning in the vision system of a demonstration-based robot, we use two methods in combination, YOLO and a colour histogram method. The multi-modal data fusion of these two methods is achieved using the IoU for the label matching phase and the weighted average for the probability combination phase. This approach proved to be more successful than using the colour histogram method alone, as it is better able to detect objects with the same colour or objects that undergo an abrupt change in illumination