The interplay between optimal estimation for improved action and optimal action for improved estimation

The perception-action loop is arguably one of the defining features of robotics: a robot needs to move for realizing a task, and its choice for a motion plan depends on its internal knowledge of the ‘world’. However, onboard sensors can typically only provide partial information about the world and, therefore, some level of reconstruction/inference/estimation is needed online for recovering any information not directly available from the raw sensor readings. One of the aspects that makes the action-perception loop scientifically very interesting (and challenging) is the tight coupling between action and perception performance: in order to successfully realize a task, the robot requires an accurate world model (which can include internal state estimation, parameter self-calibration, reconstruction of the surrounding environment, and so on). However, the chosen motion and actions can also have a strong influence on the accuracy of the estimated world model, with some motions/actions more informative than other ones. Therefore, there exists a need of finding a balance between choosing motion/action plans that are ‘optimal’ for the sake of world modeling but, at the same time, are also effective in accomplishing the tasks of interest. The aim of this workshop is to propose a round table for presenting and discussing the most recent and advanced solutions to the problem of optimal action/perception coupling (sometimes also denoted as active sensing), especially in the light of modern numerical methods (e.g., numerical optimal control, reinforcement learning) that have allowed in recent years to successfully attack complex optimal action/perception problems. The workshop will discuss practical applications of active sensing in a number of different robotic scenarios, including mobile robots, multi-robot systems and humanoids.