A virtual cattle herding game has offered researchers new insights into how people make decisions regarding movement and navigation. This unique study examined how dynamical perceptual-motor primitives (DPMPs)—basic movement models that simulate natural human behaviors—can be used to replicate human decision-making in navigation. Findings showed that a simple DPMP model was able to match nearly 80 percent of participants’ movement paths and predict their choices effectively, potentially benefiting AI and robotic navigation systems.

The research, conducted by a collaboration between Macquarie University in Australia, Scuola Superiore Meridionale, the University of Naples Federico II, the University of Bologna in Italy, and University College London, focused on real-time decision-making that mirrors everyday navigation challenges, like navigating crowded spaces or pursuing moving objects. Traditionally, navigation models rely on cognitive mapping, but this study supports a theory that human movement is less about complex planning and more about adapting to real-time influences from goals and obstacles.

Rather than mapping out surroundings in exhaustive detail, our navigation often unfolds as we go, influenced dynamically by immediate obstacles and objectives. Using DPMPs, researchers can model these intuitive responses across various activities, from reaching for an object to coordinating with multiple agents. In the study, participants controlled virtual agents in herding simulations where the DPMP model successfully replicated human-like movements and responses, offering potential applications in skill training for robotics and AI.

Participants undertook herding tasks where they were asked to move either a single cow or a group of cows into a pen. By analyzing the sequences in which players moved cows, researchers fed the data into the DPMP model to see if it could replicate human behavior. They found that three main patterns influenced players’ choices: the first target was the closest in angular distance; each subsequent choice was closest in angular distance to the previous target; and, when deciding between two cows, participants typically chose the one furthest from the pen’s center.

According to Michael J. Richardson, a professor at Macquarie University and the study’s senior author, these findings allowed the model to predict nearly 80 percent of participants’ herding choices, illustrating how humans select movement goals based on proximity and spatial strategy.

To enhance realism, the researchers designed a first-person herding game, departing from traditional aerial views in similar studies. This shift provided a more human-centered visual perspective, similar to role-playing video games, which better captures how we navigate our environment without an overhead view. Richardson emphasized that this approach marks a breakthrough in understanding how DPMPs can predict not only crowd behaviors but also individual decision-making when guiding virtual agents or robots.

This study offers a glimpse into future applications, where understanding human-like movement decisions can improve how robots and AI navigate complex, real-world scenarios.

By Impact Lab