Robustness
- Simple changes in the neural model, input model or synaptic
model require a complete rewrite of the topology.
- Moreover, we'd like to be able to plug in new behaviour and replace
old behaviour. This should require no or little modification of
the old system.
- The 2011 CABot3 had all these modules.
- The current CABot3 has NLP, vision and a planning stub.
- We'd like to have a clear interface. E.g. if these neurons
fire like X, and the synapses are like Y, then the Plan module will
work.
- We'd also like to make the modules work on their own. This requires
proper test suites.
- Lots (but far from all) of behaviour can also be managed with finite
state automata.
- You can run really big FSAs (e.g. language cascades) in parallel.
This might be an industrial use for large SpiNNaker systems.
