Decision Making, Motor Planning
KeywordsDelay Period Motor Plan Preparatory Activity Saccadic Reaction Time Perceptual Decision
“Motor planning” refers to the cognitive operations that are antecedent to the execution of a voluntary, goal-directed motor act or sequence.
As the term implies, motor planning refers to the mental operations that precede the initiation of a voluntary motor act. Imagine, for example, the simple act of reaching to pick up a glass of wine while dining at a restaurant. If you are like most people, you have just envisioned yourself making a deliberate reach to pick up the glass, presumably for the eventual purpose of taking a sip. Now, instead imagine that this same glass of wine is teetering due to an inadvertent bump of the table by a passing waiter. It is readily apparent that your reaching movement in this circumstance, though requiring the same basic excursion, would be substantively different in both its intended purpose and, accordingly, in the manner in which it is executed. Intuitively, the most likely distinction between a reach intended to take a sip of wine and one designed to prevent it from spilling is that the latter would need to be executed much more rapidly if it is to have any chance of producing a successful outcome. From the perspective of neural computation, the difference between the two movements would at first glance seem rather trivial – reaching faster simply requires that a “stronger” neural command be provided to the relevant musculature – but, in fact, deeper consideration of what would constitute an effective motor plan in each case reveals that the problem is not at all simple from a neural coding perspective.
As is generically true for any “plan,” a motor plan is a prescription for action that is informed by current circumstances and with an eye toward an expected, future outcome. If considered as a series of mental events, the “plan” then is what occurs after one has decided what to do (e.g., take a sip of wine) but before one actually does it; by definition, a plan can be canceled, thus resulting in no action. As detailed in another entry in this section (see “Perceptual Decision Making”), this depiction of mental events as discrete and serial is unlikely to have a direct correspondence in neural processing stages (see also Requin et al. 1988; Cisek and Kalaska 2010), but it is of heuristic value for considering the formulation of a motor plan as a problem of neural computation. With this in mind, it is a simple matter to consider how the motor plans for the aforementioned hypothetical scenarios must differ on the basis of presumed input and desired output. Clearly, a motor plan developed to take a taste of wine is guided by very different sources of information than that for stabilizing the unsteady glass. In the former case, the decision, and consequently the plan, to reach for the glass is informed primarily by a host of appetitive and social cues, while in the latter, visual (swaying motion of the glass) and tactile (vibration of the table) motion cues would likely predominate. In addition to receiving input from vastly disparate sources, the ensuing motor plans must evolve to produce outputs that optimize different movement parameters. Ideally, a reach to drink the wine would be graceful, or at least sufficiently precise, so as to avoid knocking over the salt shaker or dipping one’s sleeve in the olive oil intended for dipping bread. In contrast, grace may be sacrificed in the name of urgency when attempting to prevent a spill. Here, in addition to moving with greater speed, a more direct hand path to the glass (salt shaker be damned!) might be more effective. It goes without saying that if the respective motor plans are to specify movements having different speeds and trajectories, their corresponding neural correlates must carry information that can be translated into the appropriately distinct dynamical signals at the effector level (e.g., force, joint torque).
Along with providing a general sense for what a motor plan is, the juxtaposition of these real-world reaching scenarios makes a crucial point: motor planning, and the neural that implements it, represents a flexible interface that ties internal and external cues to motor effectors in a way that ensures that actions are appropriate to context. Once again, if motor planning could be thought of as a discrete computational stage, it is one that must be receptive to all manner of possible input and capable of producing a widely varied range of output – a formidable computational problem to be sure.
The preceding section provides a highly descriptive account of motor planning, making use of a concrete example to generate some intuition about its implications for behavior and the neural activity that underlies behavior. In the laboratory, neurophysiological studies of motor planning have a long history, and our understanding has relied extensively on behavioral tasks that incorporate what is known as an “instructed delay period,” sometimes simply referred to as the “delay period” or “instructed delay.” (The term “delay period” in this context should not be confused with its use to refer to a memory retention interval (e.g., see Funahashi et al. 1989). The delay period is the interval of time that elapses between two task instructions. The first instruction comes in the form of a sensory cue informing the subject about what to do and the second (the “go signal”) instructs the subject when to do it. In principle, the delay period is a window of time during which the subject can perform the necessary perceptual evaluation of the instruction and subsequently formulate aspects of the motor plan (e.g., select the appropriate response, specify metrics), all in advance of receiving the instruction to execute the action. Crucially, the time allowed for advance preparation contrasts with standard reaction time tasks, in which the go signal and the informative sensory cue are simultaneous, so perceptual judgment and motor planning processes are all completed within the reaction time interval. The temporal expansion of processing time that is afforded by the delay period has been oft exploited in neurophysiological studies of the covert cognitive processes that lead to an overt motor response.
Traditionally, single-neuron electrophysiological studies of motor planning have concentrated on the activity associated with either saccadic eye movements or reaching movements of monkeys trained to direct movements toward visual goals. Superficially at least, the neural activity recorded in association with performance of a “delayed saccade” or “delayed reaching” task is strikingly similar. Whether recorded in saccade-specific or reach-specific brain regions, a substantial fraction of neurons display two temporally discrete periods of activation, one linked to the initial stimulus indicating the movement goal (sensory event) and another to the motor response (saccade or reach). Interposed between these event-linked bursts of activity is the so-called delay period activity, a major focus of neurophysiologists’ attempts to understand how motor planning is represented in the activity of single neurons. Consistent with the logic of the instructed delay task, activity during this period demonstrates the hallmark features of that which would constitute a plan for action. First, such activity reflects the vector of the upcoming movement well in advance of the imperative signal to actually produce it (i.e., the go signal). And second, as mentioned earlier, such delay period representations of future movement need not lead to execution – plans can be canceled – as has been convincingly demonstrated with behavioral tasks that later “countermand” the initial instruction of “what to do.” That the specification of movement parameters in neural activity does not inevitably lead to the initiation of movement firmly supports the interpretation of such premovement activity as a motor plan and not a motor command.
Specificity of Motor Plans
The question of precisely what information is being represented in motor planning activity can be a difficult one, particularly so for the skeletomotor system where it is not entirely clear what movement parameters are being optimized upon execution (e.g., see Shenoy et al. 2013). For example, returning to the example of how to plan movements of different speed, one would not be particularly surprised if during movement execution, the concurrent activity of motor-related neurons displayed different firing rates in association with movements of different speed. But is there a neural correlate of the intention to move at a certain speed? More specifically, could one predict from delay period activity whether an impending movement will be fast or slow? Indeed one can, as shown in an elegant study by Churchland and colleagues (2006; see also Cisek 2006) that exemplifies the way in which different intentions may manifest as differences in neural activity. In this experiment, monkeys were trained on an instructed delay task to make either a high- or low-velocity reach depending on the color of the target to be acquired. Single-neuron recordings from dorsal premotor and primary motor cortex revealed that, along with representing the vector of the upcoming movement, the delay period activity of individual motor cortical neurons was strongly modulated by the instructed speed of the reach. Though clearly demonstrating a neural correlate of planned speed, the relationship was not a simple one; the activity of individual neurons could be positively or negatively modulated by the intention to make a faster movement, and furthermore, this representation of intended speed was shown to interact with those of ostensibly orthogonal spatial dimensions like movement distance and direction.
The Churchland et al. study is but one example taken from more than three decades of work demonstrating how movement parameters ranging from low (e.g., force) to high (e.g., trajectory) level can be observed in the preparatory activity associated with limb movements (e.g., Tanji and Evarts 1976; Wise 1985; Kalaska et al. 1989; Riehle and Requin 1989; Fu et al. 1995; Crammond and Kalaska 2000). As is the case for movement speed, it is not fully clear how the varied and sometimes redundant information concurrently present within motor preparatory activity is read out to yield the eventual motor commands. This is a rich topic and well beyond the scope of this short entry; nevertheless, it should be appreciated that current models of how plans become action are necessarily constrained by assumptions about how such signals would need to be transformed to yield the requisite downstream signals.
Planning of Eye Movements
Relative to the skeletomotor system, the oculomotor system is considerably less complex biomechanically, and as such, its study has been instrumental to deciphering how cognitive processes influence the neural activity that leads to movement. However, in contrast to reaching, introspection is not quite so valuable for considering how motor plans might differ when it comes to making saccades. Indeed, while making approximately three saccades every second, the complex decision and motor planning processes that inform one’s choices of where and when to look go largely unnoticed. Yet such decisions and the plans that embody them are essential to effective interaction with the environment, whether that entails watching a movie, hitting a baseball, or just searching for one’s keys on a cluttered countertop.
While it may be difficult to deduce from one’s own eye movements, there is no argument that cognitive strategy, or planning, plays a major role in guiding saccade choices. In his pioneering studies of nearly a half century ago, Russian psychologist Alfred Yarbus showed (Yarbus 1967) that subjects who were viewing static images could be induced to alter their patterns of saccadic landing points simply by changing task instructions (and not the image itself). Thus, the way in which saccades interrogate a visual scene depends heavily on prior knowledge of what information needs to be gleaned from it. While many subsequent insights have come from psychophysical studies of natural visual scanning, studies attempting to expose the neural correlates of visual decision making and saccadic motor planning have typically relied on more controlled task settings. As is the case for limb movements, the neural signature of oculomotor planning in its most fundamental form is that of a representation of impending saccade vector. Such is readily observed in single-neuron recordings from monkeys trained to perform variants of a delayed saccade task, and at least qualitatively similar spatial representations have been observed in visuo-oculomotor territories in frontal cortex, parietal cortex, midbrain, thalamus, and basal ganglia (e.g., Thompson et al. 1996; Snyder et al 1997; Glimcher and Sparks 1992; Schlag and Schlag-Ray 1984; Schlag-Rey and Schlag 1984; Hikosaka et al. 1989). But more than simply predicting where the eyes will go, a correlate of when the eyes will move, or saccadic reaction time can also be observed in the preparatory activity of visuo-oculomotor neurons, specifically those with ties to brainstem circuits that are proximal to the extraocular muscles (e.g., Hanes and Schall 1996; Dorris et al. 1997).
Planning Versus Executing
At what point does an oculomotor plan become a commitment to action and what is the neural basis of this transition from the merely possible to the inevitable? In a compelling demonstration, Hanes and Schall (1996) tested the hypothesis that commitment to a particular saccadic choice occurs only when the preparatory activity of oculomotor neurons representing that choice (i.e., a given saccade vector) exceeds a specific or fixed threshold. They recorded from saccade-related neurons in the primate frontal eye field while monkeys performed the simple task of looking toward a single visual target presented at an eccentric location. Importantly, even when looking toward a highly salient and unambiguous goal, saccadic reaction times are known to display considerable variance. Hanes and Schall demonstrated that the fixed-threshold mechanism could account for this variance by verifying two predictions about frontal eye field saccade-related activity. First, they observed that the activity of individual neurons increased monotonically leading up to saccade onset, and just prior to saccade onset, activity consistently attained the same level. The second prediction is a necessary consequence of the first one. If the commitment to move occurs when the increasing motor planning activity crosses a fixed threshold, then the reaction time should be inversely related to the rate at which the plan rises: shorter reaction times should correspond to motor plans that build up quickly and longer reaction times should occur when the accumulation is more gradual. Indeed, this is precisely what they observed. In an interesting variation on this experiment, Hanes and Schall also employed a countermanding version of the task wherein, on some trials, monkeys were cued to withhold the already instructed saccade. They found that trials in which the monkey was successful in withholding the eye movement corresponded to those in which the accumulating activity of frontal eye field neurons failed to achieve the putative threshold criterion – i.e., the neural correlate of a canceled motor plan.
Motor Planning Dynamics
The above-described examples from the skeletomotor and oculomotor systems convey the essence of what a motor plan is and how it is manifested in the neural activity that precedes an action. Such an account, however, fails to capture the dynamism of how motor plans are informed by perceptual events and changing internal states in the milliseconds leading up to a motor choice. As noted above, the perception-action cycle has been conceptualized as consisting of a series of discrete computational stages, a characterization that is reinforced by the way in which its neural basis has been studied. For the most part, the neural activity associated with performing a perceptually guided motor response has been examined with tasks, like the instructed delay task, that promote or even mandate serial completion of perceptual evaluation and motor planning processes. But, unlike for many laboratory tasks, perceptual information and behavioral goals are fluid in real life, and accordingly, plans may evolve in ways that reflect this ever-changing reality. As discussed at length in another entry (see “Perceptual Decision Making”), the notion that accumulating sensory evidence is manifested directly in developing motor plans (e.g., Gold and Shadlen 2000) blurs the distinction between perceptual and motor processes and hints at a more continuous interplay between perceptual evaluation and response selection. In fact, studies of perceptual decision making have consistently demonstrated that the neurons that accumulate sensory evidence in favor of a particular choice are very often the same ones that plan the action that serves as the ultimate expression of the decision/choice process (e.g., Requin et al. 1988; Gold and Shadlen 2000; de Lafuente and Romo 2006; Hernández et al. 2010). Clearly, a direct and intimate relationship between perceptual decision and motor planning makes good sense if the objective is to base actions on the most currently available information. In fact, the immediacy of this relationship is emphasized by recent studies of urgent decision making with tasks that prioritize the incorporation of incoming sensory information to guide ongoing motor plans. Remarkably, these studies suggest that motor systems may simultaneously harbor multiple and sometimes mutually exclusive motor plans until just milliseconds before a single alternative is committed to action – a conflict resolved on the basis of a timely perceptual cue (Cisek et al. 2009; Cisek and Kalaska 2010; Stanford et al. 2010). Ultimately, understanding the neural basis of motor planning will require a more complete framework for describing how perceptual and motor systems interact to affect the right action at the right time.