|
Visual feedback control for natural motion| old_uid | 14665 |
|---|
| title | Visual feedback control for natural motion |
|---|
| start_date | 2014/11/21 |
|---|
| schedule | 10h30 |
|---|
| online | no |
|---|
| summary | Sensor-guided robotic locomotion in different media often takes inspiration from natural examples. Interestingly, several naturalistic observations show that even very different species exhibit some similarities in their locomotion patterns : a notable one perhaps being the spiraling nature of paths that in some cases can be observed in sensory-guided tasks. As often conjectured in naturalistic studies, a common optimality principle may underpin such motion behaviors.
In the first part of this presentation, I will show in a robotics framework that spiraling motions appear in the solution of the problem of minimum path length. In particular, I will present an optimal solution for the visual servo control of a unicycle-like vehicle equipped with a monocular fixed vision system. The system, subject to nonholonomic constraints imposed by the vehicle kinematics and to Field-Of-View (FOV) constraints imposed by the camera, have to reach a desired position on the motion plane following the optimal (shortest) path. After showing the extremal curves, i.e. curves that satisfy necessary conditions for optimality, I will provide the partition of the motion plane in regions such that the optimal (if any exists) path from each point in that region is univocally determined. Once the optimal synthesis is available, a crucial step towards the practical application of the shortest path synthesis is to translate the optimal trajectories into feedback control laws. In this talk, I will present such feedback control laws with a sketch of the proof of asymptotic stability.
In the second part, I will show how a single camera is enough to design effective feedback control laws for a mobile robot to go through a door. The approach is derived from the natural geometry induced by the presence of a door in the environment, e.g. bundle of hyperbolae, ellipses and circles. The plane around the door is hence foliated by using confocal (the feet of the door being the foci) hyperbolae and ellipses (a.k.a. elliptic coordinates system) and confocal circles that intersect at right angles (a.k.a. bipolar coordinates system). Using visual servoing I will show that these coordinates can be directly measured in the camera image plane and feedback control laws as well as proofs of asymptotic stability of the controlled system will be provided. This work also furnishes some interesting, although preliminary, insights into how vision systems and the geometry of the environment may influence the shape of the paths in human beings while performing a rest–to–rest task. |
|---|
| responsibles | Fenouil |
|---|
| |
|