Exploring Saliency for Learning Sensory-Motor Contingencies in Loco-Manipulation Tasks.

In the first video (1.manipulation_task), we illustrate the real-world experiment described in Section 5.2 of our paper. This experiment involves a static manipulator programmed for object sorting by weight. Utilizing two differently weighted objects, we register two distinct Sensory-Motor Trace (SMTs). Initially, we register the first SMT and generate its corresponding SMCs-tree using our proposed methodology. Subsequently, the second SMT is recorded, leading to the completion of the final SMCs-tree. The video concludes with the manipulator autonomously sorting objects placed in random initial positions.

The second video (2.loco_manipulation_task), showcases the real-world experiment detailed in Section 5.3 of our paper. This experiment features a mobile manipulator programmed for object retrieval and placement tasks, involving operations such as navigating through different rooms, and opening doors and cabinets. The manipulator is tasked with retrieving objects from a wardrobe and returning them to their specific locations based on their type. We demonstrate two separate Sensory-Motor Traces (SMTs) and illustrate the resulting SMCs-tree. The video culminates with the autonomous execution of these tasks, showcasing the process by which the robot identifies and executes the appropriate SMCs in accordance with the methodology we have developed.