• No se han encontrado resultados

Modalidad de Transacción

In document ADVERTIMENT ADVERTENCIA WARNING (página 32-36)

III. CARACTERIZACIÓN DEL SECTOR INMOBILIARIO Y DE LA CONSTRUCCIÓN

III. 2. Modalidad de Transacción

Collective decision making is essentially the convergence of a group on a single choice, for example that a certain agent should move to a certain position or area. This is important for use in a nuclear cave as it will enable reduced interference and maximum efficiency for mapping. This section is split into three topic areas: consensus achievement, task allocation and collective fault detection.

2.5.3.1 Consensus Achievement

Consensus achievement is the task of reaching a single decision from several alternatives. This can be difficult since the best choice may change over time or may not be obvious to the robots with their limited sensing. Consensus achievement can be approached in two different ways. First, it can be achieved using direct communication, with each robot being capable of communicating its preferred choice, or some related information. Second is via indirect communication, where robots communicate based on some indirect cue for example population density. Consensus achievement could be useful in a heterogeneous swarm as it may help agents decide which individual is best equipped for a certain task. In the case of exploration of a nuclear cave, this could involve deciding which robot has the locomotive capabilities to reach certain areas.

Consensus achievement can be found in nature, for example in the collective decision to forage for food. This then assigns the questions of who will fill which role, and what is the shortest

path? One method to communicate this information is through trophallaxis. In nature this is the transfer of fluids from mouth to mouth, or mouth to anus. In robotics this is akin to short range communication, and has been used to find the shortest path using a PFSM in work by Gutiérrez et al. [96].

Inspiration has also been derived from the democratic nature of government. This is easy to understand as it involves the swarm voting on decisions and the decision with the most votes becoming that which is followed. This has been examined with regards to hunting, and the decision between which of two moving prey to pursue [250].

As has often proved to be the case in swarm systems, AE proves a useful tool for collective decision making. In this case though AE may be used to ascertain the optimum time to switch between behaviours. This has been researched using s-bots, implementing AE to find the optimum time to switch between group and individual behaviour [237].

Overall, whenever working with a swarm there will be some element of consensus achieve- ment, whether this be passive using inhibiting behaviours, or active, through the use of methods like voting. Usually consensus achievement requires that robots be in communication to make the best decision. This might not always be possible in a nuclear cave, and so the ‘reactive virtual forces’ framework proposed in this thesis does not directly utilise active consensus achievement.

2.5.3.2 Task Allocation

Task allocation is the process of distributing a swarm between numerous different responsibilities, in order to maximise the efficiency of the operation. The tasks to which robots are allocated are likely to change over time. Hence, dynamic task changing is useful and usually implemented with PSFMs. In general, this is particularly important for heterogeneous swarms where different robots are likely to have different capabilities and hence perform better at different tasks, for example characterising different areas within a nuclear cave environment.

As with multiple facets of swarm systems, biology offers some solutions to the challenge of task allocation. In general, social insects have several castes in order to improve efficiency. This is examined in work by Momen et al. [170]. In this work robots are divided into three castes:

• Larvae - this caste has two states with random transition: hungry or satisfied.

• Brood Carers - this caste takes food from the food dump to feed the larvae, and if the food falls below a threshold they aid in foraging.

• Foragers - this caste searches for food and drops it in a convenient location for the brood carers.

The use of these castes means that the task of feeding the larvae is completed more efficiently, and with less interference, than a homogeneous allocation of tasks. A similar method could be imagined for exploring a nuclear cave, where robots are assigned explorers and data hubs.

The explorers seek to attain new data, whilst the hubs maintain connectivity and transmit the acquired information out of the cave.

In addition to caste systems, social insects also react to stimuli that are related to the task that is to be completed; these stimuli usually have some threshold for activation. This behaviour has been ported to robotics in a study by Krieger et al. [134]. In this work a group of 12 robots was given the task of maintaining energy levels in a nest. Collecting food items increases energy in the nest, whereas foraging and moving food items decreased it. The threshold behaviour was implemented in a simple manner: when the nest energy fell to certain levels different robots would aid in foraging. It was found that the energy level was maintained by use of only this one simple activation method.

A further method for task allocation is based on the concept of bidding. In this framework robots bid on tasks based on their perceived ability to complete them. They are offered incentives in the form of virtual money, or other form of reward on completion of a task. The lowest bidder is generally awarded the task, akin to the lowest contractor being awarded a contract. An overview of these methods can be found in the survey put forward by Dias et al. [64]. A good example is the use of bidding to handle an emergency, where multiple alarms need to be switched off [161]. In this case the bids are based on the robot’s distance to the alarm, the smallest distance gets awarded the task and hence the most efficient outcome prevails.

In contrast to the bidding method is the broadcast of eligibility [249]. In this paradigm instead of bidding, the eligibility of each robot is compared and the robot with the best eligibility for a task inhibits that behaviour in other robots. This allows for dynamic task assignment due to the predisposition for eligibility to change over time. In addition, this can easily be applied to a heterogeneous swarm, whose individual eligibility may greatly differ.

In general, there is no one defined way in which tasks are assigned in swarm robotics. Later in this work, the ‘reactive virtual forces’ framework will be shown to exhibit some task allocation abilities. Robots exchange the location of areas they are unable to reach in order to recruit other members as potential solutions.

2.5.3.3 Collective Fault Detection

Collective fault detection is the ability of a swarm to identify faulty agents within the swarms’ rank. This is an important area of research as currently implemented robots are not fully reliable and hence prone to failure. In hazardous environments, such as a nuclear cave, this becomes even more pertinent.

Once again, examples have been drawn from biology to inspire swarm behaviour. Particularly in the synchronised flashes of fireflies. In fireflies this synchrony is achieved through a threshold that when reached, which causes them to flash. If nearby fireflies see this flash, then their activation threshold is increased by an amount, this eventually leads to synchrony. This can be utilised by robots to detect a faulty robot, as if a robot is not flashing in synchrony it can be

assumed to be faulty [49].

A more commonly used method is the ’I am okay’ signal [256]. This method relies on the peri- odic broadcast of a signal that informs the rest of the swarm that the broadcaster is functioning correctly. If this message is not received, then the robot is assumed to have malfunctioned. A downside of this technique is that if only the communication systems fail, the robot may still be able to perform some functions but is assumed to be entirely broken.

The utility of fault detection in a nuclear cave environment is readily apparent. Due to the hazardous nature of nuclear environments it is likely that the on-board electronics may fail. In order to continue operating at maximum efficiency this fault must then be detected by the rest of the swarm, and the failed robots’ tasks redistributed. Though fault detection is beyond the scope of this project, it is still an interesting and relevant area of study.

In document ADVERTIMENT ADVERTENCIA WARNING (página 32-36)