Dissociating Task Error and Extrinsic Reward During Implicit Sensorimotor Adaptation

The ability to learn and execute motor skills impacts both comfort and functionality of our daily lives. Yet, well-practiced movements require frequent recalibration to account for changes in both body and environment (e.g., Shadmehr et al., 2010). As an example, consider lifting a pot of coffee. If you are expecting the pot to be light in weight, you will apply little force when attempting to lift it. However, suppose it is unexpectedly heavy, resulting in a mismatch between your predicted and actual motor output. To lift the pot successfully, you need to account for the extra weight by adding force to your movement. This recalibration of your original movement plan is the result of both a voluntary and an involuntary response to the difference between predicted and actual motor output (e.g., Mazzoni & Krakauer, 2006). Researchers have termed this process sensorimotor adaptation and they have studied its importance in guiding our movements for over a century (Von Helmholtz, 1867; Krakauer et al., 2019). A popular method used to examine sensorimotor adaptation is the visuomotor rotation paradigm (e.g., Taylor & Ivry, 2011). Here, participants are often unable to see their hand as they are performing reaches to specific targets. Meanwhile, display of a digital cursor provides concurrent, visual feedback about reach trajectories. Rotating the visual feedback from the cursor causes participants to modify the direction of their reach to compensate for the discrepancy between predicted reach trajectories and visual display of the cursor (e.g., McDougle et al., 2015). In past studies, researchers have used visuomotor rotation paradigms to distinguish between explicit and implicit learning mechanisms contributing to sensorimotor adaptation (e.g., Huberdeau et al., 2015; Mazzoni & Krakauer, 2006; McDougle et al., 2016). Implicit contributions to sensorimotor adaptation have commonly been linked to subcortical error processing (Taylor et al., 2014), with cerebellar structures encoding the discrepancy between predicted motor outputs (e.g., successfully reaching to a target) and sensory feedback (e.g., perturbed cursor display) (Albus, 1971). This sensory prediction error serves as a signal to update the motor plan for subsequent movements (e.g., reaches) and their predicted outcomes, collectively referred to as the forward model (Flanagan et al., 2013; Ito, 2006). In contrast, explicit learning is often classified as a voluntary process, predominantly associated with prefrontal structures (e.g., Seidler & Noll, 2008; Taylor & Ivry, 2014). Participants can consciously apply a strategy (e.g., reaching to a nearby target) to accommodate for any target error. Target errors typically provide a vectorial error signal, indicating both the direction and magnitude of an error between an intended goal (e.g., target) and actual outcome (e.g., final cursor position) (Taylor et al., 2014). Target errors are thought to primarily impact explicit adaptation processes (e.g., Mazzoni & Krakauer, 2006). However, there is evidence that target errors also impact implicit learning (e.g., Cameron et al., 2010; Kim et al, 2019). Evidence for implicit learning in response to target errors has been shown in experiments where participants adapted their reach directions in response to ‘clamped’ feedback (Kim et al., 2019). In a “clamped” cursor feedback paradigm, a person is told to aim “normally” to a target, but the visible cursor trajectory shows a consistent bias away from the target (e.g., Morehead et al., 2017). Despite instructions to ignore the perturbed cursor and aim to the target, participants adapt to the cursor “error” by involuntarily increasing the hand angle of their reaches away from the target to compensate for the visual error (Morehead et al., 2017). In addition to sensory prediction- and target error processes at work during visuomotor adaptation, reinforcement learning processes also impact on adaptation (e.g., Izawa & Shadmer, 2011). Intrinsically rewarding properties of task success (e.g., Holland et al., 2018, Kim et al., 2019) or extrinsic performance-based rewards (e.g., Galea et al., 2015; Nikoyaan & Ahmed, 2014) impact learning through reinforcement. Such rewards (or punishments) can enhance (or slow) the rate of adaptation (e.g., Nikoyaan & Ahmed, 2014) and/or its retention (e.g., Galea et al., 2015; Shmuelof et al., 2012). Adding a reward, such as money or an exploding target in response to achievement of a goal, is thought to aid adaptation through dopaminergic reward circuitry (e.g., Taylor & Ivry, 2014). This circuitry includes structures such as the prefrontal cortex and basal ganglia and is thought to aid short-term performance and longer-term learning by encoding a binary reward signal (e.g., Izawa & Shadmer, 2011). Thus, contrary to target- and sensory prediction errors, reward signals typically do not contain information about the direction or magnitude of a movement error. The learner only gains binary knowledge about whether their movement was successful and thus will either try and repeat a prior performance or explore a new movement strategy (e.g., Pekny et al., 2015; Izawa & Shadmehr, 2011). Once a learner identifies a strategy/movement that maximizes reward, they repeat, thereby consolidating the corresponding movement. This learning process appears to be accompanied by a reward-dependent increase of movement efficiency and/or motor vigor (Zhao et al., 2024). Summerside et al. (2018) showed that participants performed reaching movements with higher movement speed and reduced reaction times to targets where a performance-contingent reward signal was expected, compared to non-rewarded targets. In the context of sensorimotor adaptation, several authors concluded that reward feedback (and reinforcement learning processes generally) contributes predominantly to explicit, strategic learning (Holland et al., 2018; Izawa & Shadmehr, 2011). Holland et al. (2018) showed that participants successfully adapted to a gradually increasing visuomotor rotation based only on binary reward feedback (i.e., a green tick for on-target reaches). When the authors interfered with explicit processing, by introducing a secondary task, participants failed to adapt, indicating a strong explicit component in reward-based adaptation (Holland et al., 2018). Nevertheless, there is evidence for neural correlates of reinforcement learning processes during visuomotor tasks in cerebellar structures, which are thought to map to implicit learning (e.g., Sendhilnathan et al., 2020; Taylor & Ivry, 2014; Thoma et al., 2008). To test how reward functions in the context of target- and sensory prediction errors, these error signals have been manipulated concurrently (e.g., Izawa & Shadmehr, 2011; Kim et al., 2019; Tsay et al., 2022). Both, reward signals and sensory prediction errors were sufficient to bring about adaptation to a cursor rotation, evidenced by an absence of difference in adapted reach directions between groups with different amounts of sensory feedback (Izawa & Shadmehr, 2011). However, proprioceptive adaptation (i.e., biases in hand estimates) was dependent on quality of sensory feedback. The authors argued that there is an interactive process between reward signals and sensory prediction errors, where the presence and strength of one signal can up- or down regulate the processing of the other (Izawa & Shadmehr, 2011). Indeed, when sensory prediction error processes were put in conflict with reward signals concerning target success, implicit adaptation processes were modulated (Kim et al., 2019). In this study, the authors manipulated target size (and hence target success/reward) in the presence of clamped cursor feedback. The amount of adaptation was reduced when feedback from the clamp showed a “successful” target hit due to a larger target, versus when it missed, due to a smaller target. However, because there was vectorized target error signal (direction and magnitude), the authors were not able to conclude that attenuation of implicit adaptation was a result of reward, rather than target error. A concern with previous research on the role of reward signals in sensorimotor adaptation is how researchers defined and implemented reward signals. While some researchers used an extrinsic reward signal such as a point system (e.g., Nikoyaan & Ahmed, 2015), others relied on an intrinsic reward signal induced via target success (e.g. Kim et al, 2019; Tsay et al., 2022). The latter conflates the contributions of two distinct signals in an adaptation paradigm: a vectorial error signal and a non-vectorial reward signal. Thus, to make inferences about the contribution of reward to the sensorimotor adaptation process, these two signals need dissociating. Here we are proposing a study aimed at dissociating target error and extrinsic reward signal effects on implicit adaptation. Implicit sensorimotor adaptation and hence sensory prediction error will be controlled through clamped cursor rotated feedback. Target error, but not reward feedback, will be manipulated via target size, such that between groups, the clamped cursor will either hit or miss the reaching target (replicating Kim et al., 2019). Some participants will also receive an extrinsic, performance-based reward signal (monetary reward) on trials where they successfully direct their physical reaches into a pre-specified reward region around the original target (independent of displayed target size). As such, the reward is given in response to not adapting to the clamped cursor. We predict that implicit adaptation, evidenced by changes in reach direction opposing the clamp, will be strongest when no reward signal is present and when the clamped cursor misses the target (i.e., “no reward + target miss”). In this condition, there will be both a sensory prediction error and target error. In contrast, implicit