An Integrated Model of Collaborative Skill Acquisition: Anticipation, Control Tuning, and Role Adoption

Abstract

We studied collaborative skill acquisition in a dynamic setting with the game Co‐op Space Fortress. While gaining expertise, the majority of subjects became increasingly consistent in the role they adopted without being able to communicate. Moreover, they acted in anticipation of the future task state. We constructed a collaborative skill acquisition model in the cognitive architecture ACT‐R that reproduced subject skill acquisition trajectory. It modeled role adoption through reinforcement learning and predictive processes through motion extrapolation and learned relevant control parameters using both a reinforcement learning procedure and a new to ACT‐R supervised learning procedure. This is the first integrated cognitive model of collaborative skill acquisition and, as such, gives us valuable insights into the multiple cognitive processes that are involved in learning to collaborate.

Periodic tapping mechanisms of skill learning in a fast-paced video game.

“…Specifically, the detection of resets led to an increase of the shooting range’s lower bound, and deflations led to a decrease of the shooting range’s upper bound. Because updates are directional, the model’s shot timing update mechanism resembles typical supervised learning algorithms (Dimov et al, 2023; Gianferrara, Betts, & Anderson, 2021). Such shot timing updates were carried out in production rules in ACT-R. See Appendix C for more technical details on the ACT-R model implementation and choice of parameters.…”

“…They are useful to experimental researchers in so far as they provide a framework for further understanding how seemingly separate neurocognitive processes are brought together to yield complex behavior. In terms of computational simulations of human skill learning, the ACT-R architecture has proved to successfully replicate key patterns of skill acquisition across various performance and motor measures in video games (Anderson et al, 2019, 2021; Dimov et al, 2023; Gianferrara, Betts, & Anderson, 2021; Moon & Anderson, 2013; Seow et al, 2021).…”

“…While motor programs are usually defined at a neurophysiological level to specify how the central nervous system drives the execution of motor actions via the peripheral nervous system (Klapp, 2010; Schmidt, 1975), production systems such as ACT-R simulate such perception-action mechanisms at a higher level and tend to be more cognitive in nature (Anderson et al, 2004). That said, existing ACT-R simulations have recently demonstrated that the architecture could also successfully learn rapid skills executed at the millisecond timescale by integrating information across cognitive, perceptual, and motor modules (Anderson et al, 2019, 2021; Dimov et al, 2023; Gianferrara, Betts, & Anderson, 2021; Seow et al, 2021).…”

“…It is known for its refined sensory-motor processors and detailed timing of human motor behavior and has provided a general framework for early applications in the realm of human computer interaction (Kieras et al, 1997). In terms of skill acquisition, ACT-R also includes a more recent Controller module which simulates the progressive tuning of sensorimotor skills whereby sensory features gradually get mapped on to specific actions via a feedback and error-based learning process (Anderson et al, 2019; Dimov et al, 2023). Applied to motor timing skills, previous research inspired by recent experimentation with the control tuning mechanism (Anderson et al, 2019; Seow et al, 2021) has found that speed-specific motor timing skills could be recalibrated across game speed settings while transferring larger scale proceduralized knowledge (Gianferrara, Betts, Bothell, & Anderson, 2021).…”

Tracking the Cognitive Band in an Open‐Ended Task

“…The video game in this paper is a 2‐player version of Space Fortress, called Co‐op Space Fortress,1 which has a recent ACT‐R model (Dimov et al., 2023). The data in this paper provide a further test of the Dimov model.…”