Analysis and synthesis of effective human-robot interaction at varying levels in control hierarchy.

摘要

Robot controller design is usually hierarchical with both high-level task and motion planning and low-level control law design. In the presented works, we investigate methods for low-level and high-level control designs to guarantee joint performance of human-robot interaction (HRI). In the first work, a low-level method using the switched linear quadratic regulator (SLQR), an optimal control policy based on a quadratic cost function, is used. By incorporating measures of robot performance and human workload, it can be determined when to utilize the human operator in a method that improves overall task performance while reducing operator workload. This method is demonstrated via simulation using the complex dynamics of an autonomous underwater vehicle (AUV), showing this method can successfully overcome such scenarios while maintaining reduced workload. An extension of this work to path planning is also presented for the purposes of obstacle avoidance with simulation showing human planning successfully guiding the AUV around obstacles to reach its goals. In the high-level approach, formal methods are applied to a scenario where an operator oversees a group of mobile robots as they navigate an unknown environment. Autonomy in this scenario uses specifications written in linear temporal logic (LTL) to conduct symbolic motion planning in a guaranteed safe, though very conservative, approach. A human operator, using gathered environmental data, is able to produce a more efficient path. To aid in task decomposition and real-time switching, a dynamic human trust model is used. Simulations are given showing the successful implementation of this method.