1 Minimalist Collective Gradient Ascent with Real Robots: Implementing Secondary Swarming Chris Melhuish, Jason Welsby Intelligent Autonomous Systems (Engineering) Laboratory Faculty of Engineering, University of the West of England Coldharbour Lane, Frenchay, Bristol BS16 1QY Phil Greenway, W.A.Wright BAE SYSTEMS (ATC) FPC 267, PO Box 5, Filton, Bristol BS34 7QW Contact: chris.melhuish@uwe.ac.uk Abstract: This paper addresses and highlights some of the problems facing designers and those who engineer small scale robots in the future. It specifically looks at the problem associated with a group of small-scale robots ascending a gradient field. In particular the performance of an individual, minimalist robot can be improved when a group of similarly limited robots is employed; being a member of a collective confers benefit to the individual. This paper reports on the implementation of earlier simulation work with a group of real ‘blimp’ robots, with a severely restricted payload, demonstrating that spatial integrity of a group of agents around a target can be improved when employing the mechanism of secondary swarming. 1. Introduction This paper describes the implementation of a set of collective minimalist sensing and control algorithms [Melhuish and Holland 1996] on a group of real autonomous robots, which are severely limited in their computation, communication and sensing capabilities. This paper reports on the utility of algorithms that have previously only been demonstrated in simulation [Melhuish, 1999]. Here it is shown that it is possible to use these ideas to control a group (or flock) of helium balloon robots which have a maximum payload of 93g and limited communications. The ability of a group of robots to reliably perform their task collectively employing limited computation, sensing and communication offers many potential advantages. For instance, a collective group has built-in redundancy, in that it can withstand a certain amount of agent loss and still achieve its goal, whereas a solitary complex robot may be disabled by the loss of a single sub-system, making the task impossible. Recently, engineers have drawn on such lessons