In ubiquitous robotic systems, sensors and actuators are distributed as modules in the environment, where these modules are able to communicate and cooperate with each other through the network. The ubiquitous robotic technology is able to compensate the insufficiency of artificial intelligence in the development of service robots. It has great advantages including low cost, easy expansion, good reusability, high efficiency and high robustness.
One of the key problems for the ubiquitous robotics is how to coordinate the heterogeneous components in different tasks, in optimization of time and resource consumptions. Compared to the task planning for traditional robotic system, the tasks in ubiquitous robotic system are of higher dimension and with nondeterministic and non-stationary nature.
To address these problems, a new model for task planning is proposed. This model has the expressivity for nondeterministic problems and the efficiency for the high dimensional problems. Based on this new model, an online adapting algorithm for the non-stationary environment is developed, aiming at establishing a ubiquitous robotic system that can adapt and learn. Further, an automated task decomposition method is developed, in order to solve the high dimensional problem. This is achieved by analyzing the variable dependencies. The similar sub-problems could be reused to further improve the efficiency. The task planning methods are tested both in simulation and physical systems.
The main contents of this study are:
A Hierarchical Option Causal Abstraction (HOCA) algorithm is proposed. The main procedures are like this. First, a causal graph is calculated based on the dependency between variables. The state space is divided into topological graph based on the causal graph. Then in each layer, the actions or abstracted options are abstracted into higher-layered options, which induce sub-goals and sub-tasks. At last the sub-tasks and the original task are solved based on semi-MDP theory. Further, the convergence, restricted completeness and optimality of the algorithm are theoretically proved. Experiments are also carried out to validate and compare the algorithms.