This is a very nice review by one of the experts in the field of action selection. I made only a few minor edits in the main text in places where I felt there was missing information and some typos. I also added some comments for possible clarification.
-- Author response: These are all good points, I made some modification to the text with respect to each and also explain below how they are dealt with in the article.
However, I would invite the author to add more discussion on three points:
1) The mechanism of recurrent reciprocal inhibition has been a major topic of discussion in the literature on selective attention. In particular, the "biased competition" model of Duncan & Desimone, and numerous variations by others (e.g. Koch), is becoming largely accepted as a very promising explanation of many phenomena of attentional modulation observed in both the dorsal and ventral streams. It seems appropriate that these models should be discussed in the section on "Recurrent reciprocal inhibition", especially given the focus on the cortex.
-- I didn't want to get to far into the literature on attention as there is another article on that which provides some discussion of attentional modulation, it also looks as though there is a planned article on biased competition model of attention although this doesn't have an author yet (perhaps you could nominate one). However, I agree that it also useful to briefly mention this topic here in relation to action selection. It actually fits nicely into the section on self-organising processing in action selection where I had a related reference from Duncan et al., however the Duncan & Desimone article is an excellent review so I have now cited that one too. I have also included a brief mention in the section on RRI as you suggest.
2) It may be useful to distinguish two kinds of selection: Selection within a given sensorimotor system (i.e. "which cup do I want to reach") versus selection across systems (i.e. "reach or run away"). Selection within a system, such as the LIP-FEF-SC circuit for eye movement, can potentially make use of simple recurrent inhibition such as Fig 1a, which can be plausibly implemented as an on-center-off-surround type architecture in the cortex. Selection across systems which subserve different behaviors would presumably not proceed in this way (or it would suffer from the connectivity explosion discussed here), but may require a central switching mechanism such as Fig 1d, perhaps in the basal ganglia. I think would be a useful distinction to make, to clarify that the various architectures of Fig 1 are not necessarily mutually exclusive.
-- I agree with regard to the different kinds of selection and that the different architectures are not exclusive. I have made the latter clear via a brief comment in the section concerning figure 1. As you point out, the distinction between kinds of selection to some degree parallels the one i make in the article regarding selection within local areas of the brain compared to that between distal areas. I have added a couple of sentences in the section on connectivity costs suggesting that these may impact more on higher-level decisions.
3) In the section "Is action selection a genuine problem", the author discusses the question of the units of selection: Are they specific acts? Entire sequences of movements? In other words, the question concerns the underlying modularity of the things that are being selected. Here, I think the dynamical models of Erlhagen & Schoner (2002) and Cisek (2006) are relevant. In these models, the individual actions among which selection operates are not pre-determined "motor plans" or specific acts for which there exists some distinct neural population. Instead, the candidate units for selection are formed on the basis of incoming information. In other words, the "units" for selection are flexibly formed on the basis of the current behavioral situation. This mechanism may be appropriate for selection-within-systems, but does not necessarily apply for selection-across-systems, which might indeed involve competition between pre-determined circuits.
-- Again this is considered, albeit briefly, in the section on 'self-organising processing in action selection' which already includes a reference to Cisek (2007). I have now included Elrhagen & Schoner here too. I have thought about merging this section with the one on 'is action selection a genuine problem' but decided against it as I prefer to leave the latter as concerned with the general issue of the decomposition of control (and not detail specific examples/theories at that point). Instead I have made more explicit the link from that paragraph to the section on self-organising processes.