M OTOR L EARNING

Humans ability to perform motor behaviours and/or adapt to the constraints of their environment often occurs through a process of trial and error learning (Wolpert, Diedrichsen, & Flanagan, 2011). Over repeated trials, individuals make comparisons between the actual and predicted outcome of an action in order to generate feedback related to their performance. This feedback can then be used in an attempt to improve accuracy on subsequent trials. For example, if an individual attempts to reach and press a light switch, but instead finds that it was out of reach, they can use this information to make sure their starting position is closer to the switch on the next trial. As a result, the individual can develop and refine an internal action model by representing associations between the motor commands that drive the limb towards a specified movement goal (i.e., light switch), the environment that they are in and the sensory (e.g., vision and proprioception) consequences of limb movement (Krakauer & Shadmehr, 2007). Moreover, by continually engaging in this process sensorimotor adaptation (Wolpert et al., 2011; Wolpert & Flanagan, 2010) can occur, reducing motor variability and increasing accuracy. Not only do these internal action models underpin the sensorimotor control processing described above but it has also been suggested that social and communicative impairments in autism may be influenced by difficulties in developing skilled behaviours (Haswell et al., 2009; Mostofsky et al., 2006; Mostofsky & Ewen, 2011). For example, Mostofsky and colleagues (2006) demonstrated that autistic children showed increased errors when performing gestures to command, gestures with imitation and gestures with tool use. Consequently, motor learning and the formation of internal action models

has been examined in autism as well as the aforementioned differences in motor behaviour and control (Fournier et al., 2010; Gowen & Hamilton, 2013; Leary &

Hill, 1996).

As described in previous sections, autistic individuals have been shown to be generally less accurate and more variable during locomotion (Calhoun et al., 2011;

Rinehart, Tonge, et al., 2006; Vernazza-Martin et al., 2005) and manual aiming movements (J. Cook et al., 2013; Glazebrook et al., 2006; Hayes et al., 2018).

However, the sensorimotor processes that underlie the formation of internal action models during sensorimotor learning seem to be operational (Gidley Larson, Bastian, Donchin, Shadmehr, & Mostofsky, 2008; Haswell et al., 2009; Hayes et al., 2018;

Izawa et al., 2012). When autistic participants’ vision was perturbed via a prism, they were able to adapt their motor output and reduce error during a ball throwing task (Gidley Larson et al., 2008). Following a baseline period where participants threw a ball to a target, they repeated the same task whilst wearing prism goggles that perturbed their vision to the right by 17°. Participants from both groups showed an immediate increase in error upon changing condition but adapted similarly, by reducing error across trials. Finally, participants returned to the non-perturbed condition and importantly, both groups demonstrated immediate after-effects (i.e., error increased). This finding shows that during the perturbed condition, both autistic and controlled participants successfully formed an internal action model that

represented the expected sensory and motor consequences associated with that condition. Therefore, when returning to the control condition, this model was no longer accurate and as a result the immediate increase in error was observed.

Furthermore, it has been shown that the learning of a three-segment movement sequence in autism is also similar to that of controls (Hayes et al., 2018). During an acquisition period, where knowledge-of-results was provided, autistic participants

modulated their motor output becoming more accurate and less variable. Importantly this adaptation was maintained in retention, providing evidence of learning. A group difference between the autism and control groups was however present throughout the study. Whether this difference was related to how participants structured the three-segment movement sequence was not examined, but the findings suggest that execution differences in autism are potentially related to issues in the sensorimotor control processes described above, and not a fundamental problem in the formation and refinement action models.

That said, how autistic individuals are able to generalise internal actions models does seem to be different (Haswell et al., 2009; Izawa et al., 2012; Marko et al., 2015; Mostofsky & Ewen, 2011; Nebel et al., 2016). The ability to generalise is reflected in how well an individual can execute an action associated with an existing internal action model under conditions (i.e., direction of movement) that differ to those in which it was developed (Shadmehr & Moussavi, 2000). As discussed in the subsection on sensory systems, autistic participants showed better performance when transferring a learned motor skill to an intrinsic coordinate where proprioceptive feedback was similar to that experienced during learning, than an extrinsic coordinate where the visual feedback was similar (Haswell et al., 2009). This

potential prioritisation of proprioceptive feedback (Haswell et al., 2009; Izawa et al., 2012) could have important implications for how autistic individuals interact with their environment given the bi-directional links between perception and action (Prinz, 1997). It is therefore of interest that Haswell and colleagues (2009) also investigated whether a relationship was present between the extent to which autistic participants prioritised proprioceptive feedback and measures of autism severity (e.g., ADOS; SRS) and imitation ability. In all cases they found that the greater an autistic child’s social impairment, the more they prioritised the proprioceptive

feedback. The implication is that although autistic participants can successfully develop new internal action models (Gidley Larson et al., 2008; Hayes et al., 2018), differences in sensorimotor integration and/or processing may occur (Haswell et al., 2009; Izawa et al., 2012). This is consistent with the finding of altered neural activity during motor learning in autism (Müller, Cauich, Rubio, Mizuno, & Courchesne, 2004; Müller, Kleinhans, Kemmotsu, Pierce, & Courchesne, 2003). For instance, Müller et al. (2004) found greater activation of the premotor cortex occurred during the later the stages of learning for autistic participants in comparison to controls, which may not necessarily support effective internal action model formation and motor performance (Müller et al., 2004). Similarly, differences in motor ability in autism have been associated with deformation of the basal ganglia (Qiu, Adler, Crocetti, Miller, & Mostofsky, 2010), an area that alongside the motor cortex

(Eliassen, Souza, & Sanes, 2001), and the cerebellum is thought to be responsible for motor learning processes (Doyon et al., 2009; Shadmehr & Krakauer, 2008).

Overall, the extant evidence indicates that autistic participants do show the ability to develop and refine new internal action models (Gidley Larson et al., 2008; Hayes et al., 2018), but they may be autism specific (J. Cook, 2016; Mostofsky & Ewen, 2011) due to differences in sensorimotor integration and/or processing (Haswell et al., 2009; Izawa et al., 2012) .