Curiosity shapes spatial exploration and cognitive map formation in humans

Active exploration of novel environments supports the formation of hippocampus-dependent cognitive maps, which are critical for navigation. A central open question is which cognitive or motivational factors drive exploration, especially when spatial knowledge is acquired without external reinforcement. Theories propose that curiosity—the desire to seek novel information—may motivate exploration and support map construction, yet this link has not been directly tested in humans. Building on frameworks distinguishing epistemic and perceptual curiosity, the authors hypothesised that higher state curiosity prior to entering a novel room would promote spatial exploration and, in turn, yield more precise cognitive maps. They designed a virtual-world exploration task where participants rated pre-room curiosity (anticipatory motivation) and post-room interest (retrospective engagement), enabling tests of their distinct relationships to spatial and visual exploration and subsequent spatial memory.

Prior work demonstrates that exploration facilitates cognitive map learning in animals and humans and implicates the hippocampus in spatial navigation. Human curiosity research has predominantly focused on semantic knowledge acquisition, with epistemic and perceptual curiosity linked to exploratory information seeking and memory enhancement. However, spatial exploratory behaviours typical of motile species and the impact of curiosity on spatial memory and cognitive map formation have been underexplored in humans. The authors position their study to fill this gap by examining how state curiosity relates to spatial (locomotor) and visual exploration and how these behaviours predict the fidelity of spatial memory representations. They also consider trait curiosity dimensions (e.g., Stress Tolerance) that may modulate the curiosity–exploration relationship.

Design: Two experiments used a desktop virtual reality exploration task comprising familiarisation, exploration (16 rooms), and an immediate sketch map memory test (Experiment 2 only). Participants navigated an outdoor pier and pathway into novel rooms and freely explored without time limits. Participants: Experiment 1: 32 recruited; final N=28 (3 men, 25 women; age 18–25, mean=19.79, SD=1.70) after exclusions (scale adjustments; incomplete exploration). Experiment 2: N=60 (5 men, 55 women; age 18–25, mean=19.6, SD=1.28). All had normal hearing and normal/corrected vision; provided informed consent; ethics approved by Cardiff University School of Psychology. Virtual environments: 18 rooms (2 for familiarisation; 16 experimental), each 16 m × 16 m. Rooms were populated with layout-defining objects (e.g., sofas, bookshelves) and smaller details; some key objects were partially occluded to require movement for complete mapping. The outdoor pier and zigzag pathway connected to each room; room order was randomised. Built in Unity 3D v2019.4.15. Apparatus: LCD monitor (1920×1080, 60 Hz); keyboard for forward/back (W/S), mouse for orientation; footsteps audio via headphones; sideward movement restricted; speed ~3.4 m/s. Position and head-direction (field-of-view angle) sampled at 60 Hz. Procedure: At the pier, participants viewed the room label, rated pre-room curiosity (1–10), walked to the room, entered (5 s animation), explored freely, then rated post-room interest (1–10) upon exit. Experiment 2 administered the Five-Dimensional Curiosity Revised (5DCR) scale before familiarisation and included a sketch map test immediately after exploration (following a 5-min break). For the memory test, participants drew room layouts on pre-outlined square sheets with the room type label; instructed to include key spatial elements (furniture, doors, windows). Drawing order was randomised. Five participants were excluded from memory analyses due to minor instruction/order changes. Exploration quantification: Roaming entropy (RE) indexed exploration variability. Path RE measured spatial coverage using Shannon entropy over accessible 0.5 m × 0.5 m grid cells (32×32, excluding occlusions) within a room; normalised by log2(k), where k is the number of accessible cells. Head-direction RE measured visual scanning using Shannon entropy over an 18×36 grid of head-direction angles (horizontal −180° to 180°, vertical −90° to 90°, 10° bins; k=648); normalised similarly. Trajectories and head-direction data were projected onto these grids. Sketch map scoring: Two independent raters scored four dimensions on 1 (Poor) to 5 (Excellent, 0.5 increments): (i) Object Presence, (ii) Spatial Distortion and Rotation of Features, (iii) Relative Positioning, (iv) Spatial Proportion. Scores emphasised spatial relationships and layout fidelity, minimising influence of artistic ability. Inter-rater calibration sessions were conducted; maps with low reliability were re-evaluated. Scoring performed via an online platform (https://map-scoring.vercel.app/; OSF: https://osf.io/s2ja7/). A composite map fidelity score was computed by averaging the four dimension scores. Inter-rater reliability coefficients ranged 0.70–0.78; Bayesian Cronbach’s Alpha posterior mean=0.93 (93% HPDI [0.92, 0.94]). Statistical analysis: Bayesian multilevel (and multivariate) models using brms (R) accounted for hierarchical data and residual correlations between outcomes. Predictors were centred around individual means to target intra-individual relationships; average values centred around the grand mean captured inter-individual effects. Priors: weakly informative normal priors centred at 0. MCMC: 4 chains, 4800 iterations per chain; convergence assessed via trace plots and Gelman–Rubin statistics; posterior predictive checks and sensitivity analyses conducted. Effects summarised by posterior means and 93% Highest Posterior Density Intervals (HPDIs). Duration spent in each room was included as a covariate (Experiment 1: M=30.35 s, SD=19.19 s; Experiment 2: M=33.10 s, SD=22.82 s).

Exploration measures varied across rooms and participants, with path RE and head-direction RE positively correlated. Curiosity and interest ratings were positively correlated. Double dissociation between curiosity and interest on exploration:

• Experiment 1 (N=28): Path RE positively associated with pre-room curiosity (β_curiosity=0.0059, 93% HPDI [0.0014, 0.0104]); tentatively negatively associated with post-room interest (β_interest=−0.0040, 93% HPDI [−0.0089, 0.0012]). Head-direction RE positively associated with post-room interest (β_interest=0.0048, 93% HPDI [0.0016, 0.0080]); indeterminate with pre-room curiosity (β_curiosity=−0.0005, 93% HPDI [−0.0030, 0.0020]). Residual correlation between RE outcomes: posterior mean=0.18 (93% HPDI [0.10, 0.27]). Differences in associations between outcomes were significant (curiosity difference=0.0064 [0.0017, 0.0110]; interest difference=−0.0088 [−0.014, −0.0030]).
• Experiment 2 (N=60): Replicated pattern. Path RE positively associated with pre-room curiosity (β_curiosity=0.0039, 93% HPDI [0.0003, 0.0073]); tentatively negatively with post-room interest (β_interest=−0.0023, 93% HPDI [−0.0062, 0.0016]). Head-direction RE positively associated with post-room interest (β_interest=0.0058, 93% HPDI [0.0034, 0.0082]); indeterminate with pre-room curiosity (β_curiosity=−0.0009, 93% HPDI [−0.0028, 0.0011]). Residual correlation: posterior mean=0.34 (93% HPDI [0.28, 0.39]). Differences: curiosity difference=0.0048 [0.0013, 0.0083]; interest difference=−0.0081 [−0.0120, −0.0041]. Effects of interest on head-direction RE remained when controlling for room type. Trait curiosity moderation (Experiment 2): Stress Tolerance strengthened the positive link between pre-room curiosity and path RE (interaction weight posterior mean=0.0041, 93% HPDI [0.0012, 0.0068]); Deprivation Sensitivity showed a potential positive effect (0.0025, 93% HPDI [−0.0003, 0.0054]); Joyous Exploration and Thrill Seeking interactions centred near zero. Cognitive map formation (Experiment 2): Composite sketch map fidelity positively associated with pre-room curiosity at the intra-individual level (β_curiosity=0.066, 93% HPDI [0.027, 0.10]); tentatively negatively associated with post-room interest (β_interest=−0.023, 93% HPDI [−0.060, 0.014]). Path RE positively associated with map fidelity (β_pathRE=0.80, 93% HPDI [0.0076, 1.59]); head-direction RE showed no intra-individual relationship (β_headRE=−0.075, 93% HPDI [−1.22, 1.12]). At the inter-individual level, average head-direction RE had weak positive evidence (β_headRE=2.22, 93% HPDI [−0.47, 4.80]). Mediation analysis suggested a modest potential mediation of curiosity’s effect on map fidelity via path RE (posterior mean=0.0028, 93% HPDI [−0.00035, 0.0077]).

The study provides direct evidence that curiosity states promote spatial exploration, supporting classic cognitive map theories. A clear double dissociation emerged: pre-room curiosity selectively predicted increased spatial coverage (path RE), while post-room interest predicted increased visual scanning (head-direction RE). This suggests distinct cognitive mechanisms: curiosity may drive movement through space to reduce uncertainty and discover new areas, whereas interest may focus attention on specific features for detailed inspection. Curiosity and exploration jointly enhanced the precision of spatial-relational memory, indicating a role for curiosity-driven exploration in cognitive map formation. Individual differences in Stress Tolerance amplified curiosity’s impact on spatial exploration, highlighting the importance of coping with uncertainty in translating curiosity into action. Potential neural mechanisms involve dopaminergic midbrain–hippocampal interactions previously linked to curiosity-enhanced memory and stabilisation of cognitive maps. Distinguishing spatial and visual exploration aligns with animal work (e.g., rearing behaviour) and suggests different processing modes contribute to spatial learning in humans.

States of curiosity enhance spatial exploration and improve cognitive map formation in humans, providing the first direct evidence for a core proposition of cognitive map theory. The work clarifies distinct roles of curiosity (locomotor exploration) and interest (visual exploration) and shows that trait Stress Tolerance strengthens curiosity-driven exploration. These insights have practical implications for the design of buildings, urban spaces, museums, and virtual environments to harness curiosity for improved exploration and memory. Future research should probe neural mechanisms (dopaminergic–hippocampal circuits), test generalisability across populations and environments, integrate complementary navigation tasks, and dissect how interest may differentially support item-specific versus relational spatial memory.

• Cognitive map assessment relied primarily on sketch maps; complementary navigation/wayfinding tasks could provide broader validation of spatial knowledge and its application.
• Path roaming entropy predominantly indexes spatial exploration but may also reflect aspects of visual exploration; head-direction roaming entropy mainly captures visual exploration but can be influenced by spatial layout. Despite overlaps, most variance maps onto spatial vs visual exploration respectively.
• Sample demographics were heavily skewed toward women; future studies should balance gender to assess generalisability.
• Environmental feature analysis (e.g., object novelty, arrangement) and their relationships to interest and exploration were not exhaustively modelled and represent an avenue for future work.