Ruaridh Clark

Autonomous and scalable control for remote inspection with multiple aerial vehicles

A novel approach to the autonomous generation of trajectories for multiple aerial vehicles is presented, whereby an artificial kinematic field provides autonomous control in a distributed and highly scalable manner. The kinematic field is generated relative to a central target and is modified when a vehicle is in close proximity of another to avoid collisions. This control scheme is then applied to the mock visual inspection of a nuclear intermediate level waste storage drum. The inspection is completed using two commercially available quadcopters, in a laboratory environment, with the acquired visual inspection data processed and photogrammetrically meshed to generate a three-dimensional surface-meshed model of the drum. This paper contributes to the field of multi-agent coverage path planning for structural inspection and provides experimental validation of the control and inspection results.

Consensus speed maximisation in engineered swarms with autocratic leaders

Control of a large engineered swarm can be achieved by influencing key agents within the swarm. The swarm can rely on its communication network to spread the external perturbation and transition to a new state when all agents reach a consensus. Maximising this consensus speed is a vital design parameter when fast response is desirable. The systems analysed consist of N interacting agents that have the same number of outward, observing, connections that follow k-nearest neighbour rules and are represented by a directed graph Laplacian. The spectral properties of this graph are exploited to identify leaders with a newly presented semi-analytical approach referred to as the Leaders of Influence (LoI) method. This method is demonstrated on k-NNR graphs for a set number of leaders. These methods are compared with a genetic algorithm and are shown to be efficient and effective at leader identification. A focus of this work is the effect of leadership style on consensus speed where an autocratic approach (leaders that are not influenced by other nodes in the graph) is shown to always produce faster consensus than a democratic leadership model.

Consensus speed optimisation with finite leadership perturbation in k-nearest neighbour networks

Near-optimal convergence speeds are found for perturbed networked systems, with N interacting agents that conform to k-nearest neighbour (k-NNR) connection rules, by allocating a finite leadership resource amongst selected nodes. These nodes continue averaging their state with that of their neighbours while being provided with the resources to drive the network to a new state. Such systems are represented by a directed graph Laplacian with two newly presented semi-analytical approaches used to maximise the consensus speed. The two methods developed typically produce near-optimal results and are highly efficient when compared with conventional numerical optimisation, where the asymptotic computational complexity is O(n3) and O(n4) respectively. The upper limit for the convergence speed of a perturbed k-NNR network is identified as the largest element of the first left eigenvector (FLE) of a graphs adjacency matrix. The first semi-analytical method exploits this knowledge by distributing leadership resources amongst the most prominent nodes highlighted by this FLE. The second method relies on the FLEs of manipulated versions of the adjacency matrix to expose different communities of influential nodes. These are shown to correspond with the communities found by the Leicht-Newman detection algorithm, with this method enabling optimal leadership selection even in low outdegree (<; 12 connections) graphs, where the first semi-analytical method is less effective.