Learner-Centred, Teacher-Centred and Blended Curriculum Design in Swarm Systems

Abstract Robot swarms have been used in various civilian and military applications, from entertainment to serious missions. Complex swarm tasks involve multiple interdependent skills that swarm members need to possess for successful operation. Swarm operation environments can be dynamic, noisy or with limited communication resources, which exacerbates task complexity. The non-linear interactions between swarm members make it challenging for experts to formulate agent-level rules that result in the emergence of a desirable swarm-level behaviour. Until now, there has been no general methodology that algorithm designers can follow to go from the individuals to the group behaviour. Automated swarm behaviour design has demonstrated its potential in enabling the learning of several swarm tasks. Nonetheless, incorporating human knowledge in these automated techniques is instrumental in accelerating the learning process and ensuring that the learning results in the intended behaviours. The inclusion of domain expertise in automated swarm design has mostly been conducted in an ad hoc manner without sufficiently studying how to maximise the benefit of domain expertise while not overwhelming the human expert. This chapter investigates the use of machine education as a holistic approach for teaching swarm members the required skills in complex swarm tasks. Specifically, we study various ways of designing a curriculum to systematise the process of learning and the incorporation of expert knowledge. We focus on three curriculum design approaches: learner centred, teacher centred and blended. We present case studies for each of these approaches and present the key lessons learnt and recommendations for future studies.