Model predictive motion planning for robotic recording of human activities.
Today, a wide variety of application domains require authentic data of human activities and the associated changes to the environment to enable new data-based approaches to imitation or action recognition. However, recording highly specific activities is complex and costly. A robotic observation of the environmental state and recording of the activity offers an economical alternative.
To achieve this objective, model predictive motion planning for a collaborative robot with camera is developed considering the following aspects:
- Predictive avoidance of collisions with humans to be observed in the workspace.
- Real-time capable modeling of the environment and maximization of information gain through optimal camera movement
- Seamless recording of the activity through proactive avoidance of occlusions