Multi-platform swarm/cluster missions are an attractive prospect for improved science return as they provide a natural capability for temporal, spatial and signal separation with further engineering and economic advantages. As spacecraft numbers increase and/or the round-trip communications delay from Earth lengthens, the traditional "remotely-controlled" approach begins to break down. It is therefore essential to push the management of the spacecraft into the space segment. In other words, to make spacecraft more autonomous - if the desired goal of missions involving satellite swarms is to be realised. An autonomous group of spacecraft requires coordination, but standard terrestrial paradigms such as negotiation, require high levels of inter-spacecraft communication and on-board computation power, both of which are nontrivial in space (especially in the context of smaller nanosatellites platforms). This research therefore introduces the principles of stigmergy as a novel method for coordinating a cluster. Stigmergy is an agent-based, behavioural approach that allows for infrequent communication with decisions based on local information. Supervisors/ ground stations occasionally adjust parameters and disseminate a common feedforward and feedback environment that is used for local decisions and indirect coordination. Such an autonomous group of spacecraft can be considered as an emergent system, with the top-level group behaviour emerging from the local low-level behaviours of the individuals. Analysis is presented for a number of scenarios with performance evaluated in terms of intuitive behaviours: greedy, considerate, proactive and obstinate, which are mathematically defined. This reveals that effectiveness can be steadily improved by making the behaviours more "self-aware". The inherent suitability of the stigmergy solution for spacecraft swarm coordination is demonstrated by considering its three major benefits the first of which is scalability. Stigmergy is the mechanism used to coordinate hundreds of thousands of individuals in insect colonies. Hence it can cope in missions involving large numbers of spacecraft (demonstration of its application to up to 18 spacecraft in this work) unlike other distributed planning approaches that consider only a handful. Moreover the system is hierarchical which further helps to free the ground station from the micromanagement of individual spacecraft that is essential for remote cluster missions. All this is achievable without direct coordination, which would require the need for intersatellite links, that may well be costly on nanospacecraft platforms. The second major benefit is the ability to cope with dynamic problems. Large numbers of spacecraft will always introduce some degree of chaos and dynamism to the problem. lbis is exacerbated by the fact that tasks and/or spacecraft may fail at any time. The current planning approaches would require replanning to alleviate this, which can be highly costly and slow to respond. This work shows that the self organisation of the system brings inherent resilience. Spacecraft can then easily cope with uncertainty; can reconfigure based on spacecraft failure; can cope with changing mission goals and bursts of unexpected tasks into the system - all without additional overhead. The third major benefit is that only simple algorithms are needed which is essential for low-power nanospacecraft. In essence, the spacecraft needs only to maintain a local behaviour without interacting with others. In this thesis we describe in detail exactly what an on board "behaviour" is and how evaluation of its appropriateness for different goals is studied. This continuous search for a changing optimal behaviour is performed using a genetic algorithm, which is a challenging problem posing difficulties for existing operators. The search problem is explored analytically using Markov chains, leading to the development of a new family of distribution replacement operators. These operators have the unique ability to explicitly (rather than probabilistically) control the population diversity in fitness (rather than genome) space, which translates into improved performance. Finally this thesis concludes by considering three case studies. The first shows the implementation of the algorithms on realistic nanospacecraft flight hardware, demonstrating that the system could feasibly be deployed, even with currently available nanospacecraft platform technologies. The second demonstrates hierarchical load balancing of the cluster to maintain even power usage. Finally the third case study shows how the system can operate well on more realistic multidimensional resource spaces.