An Idiotypic Immune Network As a Short-Term Learning Architecture for Mobile Robots

Abstract

Abstract.A combined Short-Term Learning (STL) and Long-Term Learning (LTL) approach to solving mobile robot navigation problems is presented and tested in both real and simulated environments. The LTL consists of rapid simulations that use a Genetic Algorithm to derive diverse sets of behaviours. These sets are then transferred to an idiotypic Artificial Immune System (AIS), which forms the STL phase, and the system is said to be seeded. The combined LTL-STL approach is compared with using STL only, and with using a handdesigned controller. In addition, the STL phase is tested when the idiotypic mechanism is turned off. The results provide substantial evidence that the best option is the seeded idiotypic system, i.e. the architecture that merges LTL with an idiotypic AIS for the STL. They also show that structurally different environments can be used for the two phases without compromising transferability.

Real-world transfer of evolved artificial immune system behaviours between small and large scale robotic platforms

“…Reference [9] is also able to demonstrate a significantly higher level of idiotypic adaptability when the mean e value is about 20%, even though it uses different robots, a slightly different architecture, different control parameters, and different test problems (although e showed a much higher standard deviation than in [8] for a given set of parameters). Prior work thus shows that e is the important parameter for achieving idiotypic stability and that as long as other parameter choices However, with all of the test problems described here a mean e value of 20% proves difficult to achieve by manipulation of the parameters k1, k2, b and , and as with [9], e shows a high standard deviation for a given set of parameters. In addition, a series of preliminary experiments demonstrates that e values less than about 35% do not appear to work well, and that, for these problems at least, high e values of about 75±10% produce better results.…”

“…However, the selected antibody is not that with the highest new affinity value; it is the matching antibody with the highest concentration. Each antibody begins with a number of clones Nt0 = 1000 (the initial value is based on recommendations in [9]), which fluctuates with time according to a variation of Farmer's equation [16]: (5) where ( N im)t represents the number of clones of each antibody matching the invading antigen m, ( P im)2 is the newly-adjusted paratope value of each of these antibodies (representing the new strength of match to m), b is a scaling constant and k3 is an antibody death rate constant. The concentration C ij of each antibody in the system consequently changes according to:…”

“…The parameters chosen are b = 200, k1 = 0.30, k2 = 1.85, and = 25. In addition, the parameter k3 is set at 0 (as in [8] and [9]) as a fixed antibody death rate is not required.…”

“…Previous papers have provided evidence that an idiotypic AIS architecture has advantages (including greater adaptability) over a reinforcement learning (RL) scheme when applied to mobile robot navigation problems [8], and have shown that the idiotypic LTL-STL architecture permits transference from the simulator to a number of different real-world environments [9] using small epuck robots, see Figure 1. The chief aim of this paper is, therefore, to supply further support for the robustness, scalability and adaptability of the architecture by showing that it can be extended to the much larger-scale Pioneer P3-DX robots, see Figure 2.…”

A Lifelong Learning Hyper-heuristic Method for Bin Packing

“…Early work relied on hand-coded antibodies within the network, and focused on learning connections. This was significantly improved by Whitbrook et al (2007Whitbrook et al ( , 2008Whitbrook et al ( , 2010, who used an evolutionary algorithm in a separate learning phase to produce antibodies that are used to seed the network. Even taking this into account, the network does not have structural plasticity-the initial learning phase happens only once and hence the networks cannot adapt to significant environmental changes.…”