The author posits that there are three cardinal points of discussion that this research has provoked, these are:
• The potential benefits of heterogeneity for exploration • The utility of virtual forces for exploration and mapping • The advantages of robotics in the nuclear industry
In this section, each of these topics will be discussed drawing on information gained from the research conducted within this thesis.
Benefits of Heterogeneity for Exploration - It has been shown that heterogeneity provides
diverse sensing and locomotive capabilities and can outperform a homogeneous swarm in ex- ploration due to its intrinsic asymmetry; these benefits are discussed in this section. However,
these benefits come with a drawback: the swarm has more capabilities spread across more agents, but if one fails there are less agents with the same capabilities to replace it. This makes a heterogeneous swarm inherently less robust. This can be accounted for by having multiple agents with the same capabilities so that a single robot failure does not significantly reduce the swarm’s performance. In the author’s view, the benefits of increased sensing, locomotive and explorative capabilities outweigh the reduced robustness.
Chapter three showed that to gather the necessary data within a nuclear cave, multiple sensing modalities would be required. A single robot would not be capable of carrying the diverse range of sensors due to size restriction imposed by entry into the nuclear cave, thus a heterogeneous swarm is required. Though heterogeneity is a necessity in this case, it is also a benefit; if a robot has a specialised sensing modality, it can examine areas of the cave that may not be of interest for other robots, while they continue to explore areas that their sensory capabilities are specialised for.
The benefit of heterogeneous locomotion was made apparent in chapter four; if a single locomotion method was used, robots would not be capable of surmounting all obstacles within a nuclear cave. Therefore, a heterogeneous approach to locomotion allows a swarm to maximise coverage of a nuclear cave. Combining locomotion methods and implementing supplementary modalities, such as a detachable grappling hook, would enable a full view of the nuclear cave. An interesting point to note is that when deciding on locomotion strategies for a swarm, the choice is usually left to the designers ‘expert knowledge’. Instead, it is possible to use comparative experiments to empirically decide the most suitable locomotion strategies once appropriate metrics are selected. This allows the benefits of heterogeneous locomotion to be fully realised.
A combination of heterogeneity in sensing and locomotion could allow robotic agents to increase their data gathering efficiency. This would encourage locomotive and sensor pairings that are of most benefit to the exploration effort. As an example, a flying robot utilising a thermal contact sensor, is likely to be unable to adequately maintain contact. However, if this sensor was placed on a wall climbing robot that could remain static during inspection, the temperature of the same area could be analysed with greater accuracy.
During the examination of the ‘Reactive Virtual Forces’ framework it was found that the heterogeneous swarm performed a more efficient exploration of the environment than its ho- mogeneous counterpart. This was discovered to be due to the intrinsic asymmetry imposed by the heterogeneous swarm. This asymmetry led to robots exploring different regions of the map and communicating their maps at different times. As the asymmetry of a heterogeneous swarm is implicit, this benefit does not need to be specifically designed and can hold true across many implementations. Exploration was achieved using the same control architecture for both heterogeneous and homogeneous swarms, which also suggests that the heterogeneous swarm did not require a more complex control strategy.
be gathered and more terrain to be traversed. However, this diversity comes with a drawback; if there is only one robot with a specialism and it becomes incapacitated, then the swarm can no longer gather data, or move to areas, associated with that specialism. This problem is reduced as the size of the swarm increases, as there are more agents that are alike.
Utility of Virtual Forces for Exploration and Mapping - The utility of virtual potential fields in exploration and mapping was shown. ‘Reactive Virtual Forces’ were used to control a swarm of heterogeneous and homogeneous robots in the task of exploration and mapping. This represents a novel investigation into control of a heterogeneous swarm using virtual potential fields. The drawback of the ‘Reactive Virtual Forces’ framework is that currently it may only be used to generate geometric maps of an environment. Though this is the most important information regarding a nuclear cave, this might not be the case for exploration of other unknown environments.
Previous work has mostly implemented virtual potential field for pattern formation, path planning and spatial distribution [251] [19] [274] [223] [199] [124]. Work that has examined exploration with virtual potential fields has focussed on the use of homogeneous swarms [155]. The ‘Reactive Virtual Forces’ framework represents a novel study into the use of virtual fields for the guidance of a heterogeneous swarm of autonomous robots in the task of exploration and mapping.
The ‘Reactive Virtual Forces’ has shown that simultaneous localisation and mapping of an unknown environment is possible using virtual fields. This is achieved solely through distance measurements and odometry. Thus, it allows for a homogeneous swarm and a heterogeneous swarm to be controlled using the same architecture, with the only changes being some virtual parameters. Virtual potential fields have shown themselves in this thesis to be a powerful tool for exploration, especially when utilised with a heterogeneous swarm.
Exploration in this case has been achieved using analogues of three forces: the gravitational force, the electrostatic force and the strong nuclear force. This shows that complex behaviour is possible through the implementation of simple rules on multiple interacting physically in- stantiated agents. An interesting point for discussion is that the implementation of more forces, or alteration of the interaction between forces, could allow for more complex behaviours to be generated. Robots could be assigned more virtual parameters to define heterogeneity and thus increase their specialisation. Though this might impact the complexity of the control architecture, it may also allow for increased performance in exploration.
The ‘Reactive Virtual Forces’ control architecture currently enables efficient exploration and geometric mapping. However, it does not allow for other metrics to be characterised. Geometric information is the foundation of mapping and is important within a nuclear cave, however in the future it would be useful to fuse other sensor data into the map. Additionally, the ‘Reactive Virtual Forces’ requires that robots utilise accurate range finding devices and does not allow for mapping via imaging. Though this is not a problem for exploration of a nuclear cave due to its
dark nature, if the ‘Reactive Virtual Forces’ framework were to be used for other exploration efforts it would be necessary to address this.
The Advantages of Robotics in the Nuclear Industry - The nuclear industry could benefit
from the implementation of more autonomous robotic systems. This would enable exploration of environments that are currently inaccessible, such as a nuclear cave, whilst also increasing plant worker safety by reducing exposure to harmful conditions. In addition, it would allow for the ongoing costs of decommissioning to be reduced, as robotic systems can work more efficiently and for longer hours than a team of plant workers.
To date there are very few robotic systems that are implemented in the nuclear sector. However, it seems they could be of great benefit. Nuclear environments present dangerous surroundings for plant workers. The radiation, temperature and fatigue that are presented can often mean that workers must work in short shifts to prevent prolonged exposure to these conditions. If an autonomous robotic system were introduced, workers could avoid these dangers and productivity could be increased.
This is exemplified by a heterogeneous swarm of autonomous robots for the remote charac- terisation of a nuclear cave environment. Such a swarm allows exploration of an environment that is not accessible to human workers due to the adverse conditions. If the same swarm were implemented in an area that workers could enter, it would bring about other benefits. One of these is reduced fatigue. Robots do not tire and can perform tasks that could become exhausting for a human worker, such as examination of an environment whilst wearing a fully protective suit. This would enable a single plant worker observing a robotic swarm to work longer hours and likely gather more accurate data. Additionally, the worker is kept away from the harmful environment if the robots are operated remotely.
To enable current plant workers to operate a robotic swarm, the user interface should be simple or familiar. This would allow workers to interact with a swarm without having specialist training. If a swarm were fully autonomous, a worker does not need to interact with it, instead they may act as a supervisor enabling intervention if a particular event requires it.