Performance optimisation of mobile robots in dynamic environments

Abstract

Rousing applications of robot teams abound over the past three decades, but ferocious demands for viable systems to coordinate teams of mobile robots in dynamic environments still linger on.

To meet this challenge, this project proposes a performance optimisation system for mobile robots to make the team performance more reliable and efficient in dynamic environments. A wide range of applications will benefit from the system, such as logistics, military, and disaster rescue.

The performance optimisation system comprises three main modules: (1) a task allocation module to assign tasks to robots, (2) a motion planning module to navigate robots, and (3) a graphical simulation module to visualise robot operations and to validate the methodologies of performance optimisation.

The task allocation module features a closed-loop bid adjustment mechanism for auctioning tasks to capable robots. Unlike most traditional open-looped methods, each of the robots evaluates its own performance after completing a task as feedback correction to improve its future bid prices of similar tasks. Moreover, a series of adjustments are weighed and averaged to damp out drastic deviations due to operational uncertainties. As such, the accuracy of bid prices is improved, and tasks are more likely allocated to suitable robots that are expected to perform better by offering more reliable bids.

The motion planning module is bio-inspired intelligent, characterised by detection of imminent neighbours and design flexibility of virtual forces to enhance the responsiveness of robot motions. Firstly, while similar methods unnecessarily entail each robot to consider all the neighbours, the detection of imminent neighbours instead enables each robot to mimic creatures to identify and only consider imminent neighbours which pose collision dangers. Hence, redundant computations are reduced and undesirable robot movements eliminated. Secondly, to imitate the responsive motion behaviours of creatures, a virtual force method is adopted. It composes virtual attractive forces that drive the robots towards their targets and, simultaneously, exerts virtual repulsive forces to steer the robots away from one another. To enhance the design flexibility of the virtual forces, a twosection function and, more significantly, a spline-based method are proposed. The shapes of force curves can be flexibly designed and adjusted to generate smooth forces with desirable magnitudes. Accordingly, robot motions are streamlined and likelihood of robot collisions reduced. The graphical simulation module simulates and visualises robot team operations, and validates the proposed methodologies. It effectively emulates the operational scenarios and enables engineers to tackle downstream problems earlier in the design cycle. Furthermore, time and costs of robotic system development in the simulation module are considerably cut, compared with a physical counterpart.

The performance optimisation system is indeed viable in improving the operational safety and efficiency of robot teams in dynamic environments. It has substantially pushed the frontiers of this field, and may be adapted as an intelligent control software system for practical operations of physical robot teams to benefit various applications.