Abstract
The performance guarantee of autonomous robotic missions is yet rarely addressed.This work follows the work proposed in previous PhD which quantitatively deals with the safety, duration and energy dimensions of the performance of planned missions in known environments. The PANORAMA (SED) (Performance and AutoNOmy using Resources Allocation MAnagement) methodology proposed manages dynamically the available hardware and software resources to guarantee, in real time, the performance objectives. The mission is split into a succession of independent activities with constant constraints where the resources to be employed are identified and configured. The localization dimension is only considered qualitatively and energy issue is adressed by minimizing the energy margin.In this document, the integration of the qualitative consideration of the localization dimension of performance is based on the construction of predictive uncertainty models of pose, largely established empirically, based on an interval approach. The proposed methodology is deployed on local localization methods using odometry with an exteroceptive resetting using Kinect sensors and Aruco markers. It also implements on a new global localization method based on appearances called LZA. It is based on the correlation between the current exteroceptive signature of the robot and those, previously calculated, within a grid covering the environment.Taking into account the localization performance qualitatively shows that a correlation can exist between 2 successive activities due to the initialization times required for the operationalization of a new localization method. This new representation of the data coupled with predictive pose uncertainty models makes possible to use the Viterbi algorithm to select the localization methods (hardware and software resources) to be implemented to satisfy the defined pose constraints. It makes also possible to address the energy performance to maximize the associated margin. A new PANORAMA (SLED) methodology enriching PANORAMA (SED) is therefore proposed.The PANORAMA (SLED) methodology is deployed experimentally on a Patrol mission executed in an indoor environment with strong perceptual aliasing. It has shown its ability to guarantee the performance objectives Security - Energy - Location - Duration. The main shortcoming comes from the fact that presently we do not consider the localization uncertainty when switching between activities.Even if PANORAMA (SLED) still has some limitations, it should, in the coming years, be extended to Exploration missions, allowing objective optimization of the discovery phase, while ensuring the recovery phase of the autonomous robot.