Reorienting in virtual 3D environments: do adult humans use principal axes, medial axes or local geometry?

Abstract

Studies have shown that animals, including humans, use the geometric properties of environments to orient. It has been proposed that orientation is accomplished primarily by encoding the principal axes (i.e., global geometry) of an environment. However, recent research has shown that animals use local information such as wall length and corner angles as well as local shape parameters (i.e., medial axes) to orient. The goal of the current study was to determine whether adult humans reorient according to global geometry based on principal axes or whether reliance is on local geometry such as wall length and sense information or medial axes. Using a virtual environment task, participants were trained to select a response box located at one of two geometrically identical corners within a featureless rectangular-shaped environment. Participants were subsequently tested in a transformed L-shaped environment that allowed for a dissociation of strategies based on principal axes, medial axes and local geometry. Results showed that participants relied primarily on a medial axes strategy to reorient in the L-shaped test environment. Importantly, the search behaviour of participants could not be explained by a principal axes-based strategy.

Figure 1. Schematics showing the geometric properties of a rectangular space.

A) The major principal axis, based on the global shape of a space, passes lengthwise through the center of a space. The minor principal axis lies perpendicular to the major principal axis. Based on the principal axes alone, the shape of a space cannot be determined. B) Absolute metrics are the precise metrics of a given area. For instance, a rectangular space is composed of a set of opposite walls that are 200 cm in length and another set of opposite walls that are 100 cm in length. C) Relative metrics are the relational properties of a given area. For instance, a rectangular space is composed of a set of opposite walls that are longer and another set of opposite walls that are shorter in comparative length. D) The medial axes are based on local shape parameters; the trunk axis passes through the length of the space and the branch axes extend to the vertices of the space. Based on the medial axes alone, the shape of an area can be determined.

Figure 2. Predicted encoding strategies.

As an example, each panel shows the corners that are predicted to be selected by group AC. A) A schematic representation of the training environment. The filled squares in the corners of the rectangle represent the response boxes that participants in group AC were trained to select. B–F) Schematic representations of the L-shaped environment. The filled squares in the corners of the L-shaped environment represent the response boxes that participants in group AC are predicted to select according to a given strategy. The open squares in the L-shaped environment represent the response boxes that participants should not select according to a given strategy. The dashed lines indicate the location of the major principal axis (B–D) and the medial axes (E).

Figure 3. Training Environment.

A) Schematic representation of the virtual training environment. Participants in Group AC were trained to choose either of the response boxes (indicated by the filled squares) located at the geometrically identical corners A and C whereas participants in Group BD were trained to choose either of the responses boxes (indicated by the open squares) located at corners B and D. B) One example viewpoint of the virtual training environment.

Figure 4. Testing Environment.

A) Schematic representation of the virtual L-shaped testing environment. The open squares represent the response boxes located at each corner. B) One example viewpoint of the virtual testing environment shown from corner U.

Figure 5. Performance during testing in the L-shaped arena.

The proportion of choices (mean and standard error of the mean) made to each of the corners in the L-shaped environment across all testing trials.