Adaptive multi-robot control through on-line parameter identification at system level
- Autores: Pragna Das
- Directores de la Tesis: Lluís Ribas Xirgo (dir. tes.)
- Lectura: En la Universitat Autònoma de Barcelona ( España ) en 2018
- Idioma: español
- Tribunal Calificador de la Tesis: Antoni Grau Saldes (presid.), Joan Oliver Malagelada (secret.), Luís Domingos Ferreira Ribeiro (voc.)
- Programa de doctorado: Programa de Doctorado en Ingeniería Electrónica y de Telecomunicación por la Universidad Autónoma de Barcelona
- Materias:
- Enlaces
- Resumen
Teams of automatic guided vehicles (AGVs) populate industrial and warehousing facilities and take care of the internal logistics. State-of-the-art, implemented solutions for decision-taking processes in planning and control of such kind of multi-robot systems (MRSs) do not encompass real-time analysis of behaviors of robots and environment. Conditions of parts of the mobile robots (MRs), state of charge of batteries and floor conditions change and have strong influence on behaviors. This work proposes a parameterized behavioral model which takes these factors into account and accurately estimates transportation costs over time. By using it, computation of path travel times gives results closer to reality than by other methods based on weighted lengths, thus helping to take better decisions for system control and management.
In this work, we have considered a plant model that resembles that of a factory, with several blocks (storage units, machinery and so on) that define the roads and crossings where MRs move. In this model, a graph, with nodes being ports (for loading, unloading, load checking, et cetera) or junctions and bifurcation and edges being the connections among them, represents the traffic network. The robots carry out the task of traversing edges to carry materials. The travel time of edges by MRs is proposed as one such cost parameter. A bi-linear state dependent model has been devised for real-time prediction of travel times. The travel times are estimated online using this model through Kalman filtering. Experiments show that average total path costs of paths obtained through on-line estimated travel times are 15% less that of paths obtained by heuristics costs.
Nevertheless, a good estimation of travel times requires historical data, obtained at close instances, but there are situations when travel times for one or more edges for the entire duration of operation are not available to an individual robot. The proclivity of this occurrence lies in the fact that edge may not have been traveled even once by the robot, or travel time for that edge have been recorded not in recent past. Then, it is imperative for that robot to gather the necessary travel times from others in the system as a reference observation, but these observations are from other robots in different battery condition than itself. Still, the model can predict travel time for the robot using other robots’ observation and its own change or exploration in the travel times until the current instance. The crux of this process is to predict current travel times in the robot using others’ travel time for the same edge. The mechanism of information sharing between one robot to others in the system has been devised in a form of a common ontology-based knowledge. This ontology includes information about edge travel times with contexts attached to each instance and helps to fetch and share information among MRs. This greatly helps MRs to estimate travel times more accurately. This affects route planning to find paths with lesser total path cost. The average of total cost of 100 paths generated through travel times obtained with sharing is 40% less than that of paths generated through travel times without sharing.
