Changes in movement patterns in relation to sun conditions and spatial scales in wild western gorillas

For most primates living in tropical forests, food resources occur in patchworks of different habitats that vary seasonally in quality and quantity. Efficient navigation (i.e., spatial memory-based orientation) towards profitable food patches should enhance their foraging success. The mechanisms underpinning primate navigating ability remain nonetheless mostly unknown. Using GPS long-term tracking (596 days) of one group of wild western lowland gorillas (Gorilla gorilla gorilla), we investigated their ability to navigate at long distances, and tested for how the sun was used to navigate at any scale by improving landmark visibility and/or by acting as a compass. Long episodic movements ending at a distant swamp, a unique place in the home range where gorillas could find mineral-rich aquatic plants, were straighter and faster than their everyday foraging movements relying on spatial memory. This suggests intentional targeting of the swamp based on long-distance navigation skills, which can thus be efficient over a couple of kilometres. Interestingly, for both long-distance movements towards the swamp and everyday foraging movements, gorillas moved straighter under sunlight conditions even under a dense vegetation cover. By contrast, movement straightness was not markedly different when the sun elevation was low (the sun azimuth then being potentially usable as a compass) or high (so providing no directional information) and the sky was clear or overcast. This suggests that gorillas navigate their home range by relying on visual place recognition but do not use the sun azimuth as a compass. Like humans, who rely heavily on vision to navigate, gorillas should benefit from better lighting to help them identify landmarks as they move through shady forests. This study uncovers a neglected aspect of primate navigation. Spatial memory and vision might have played an important role in the evolutionary success of diurnal primate lineages.

Bio-inspired multi-scale fusion

We reveal how implementing the homogeneous, multi-scale mapping frameworks observed in the mammalian brain's mapping systems radically improves the performance of a range of current robotic localization techniques. Roboticists have developed a range of predominantly single- or dual-scale heterogeneous mapping approaches (typically locally metric and globally topological) that starkly contrast with neural encoding of space in mammalian brains: a multi-scale map underpinned by spatially responsive cells like the grid cells found in the rodent entorhinal cortex. Yet the full benefits of a homogeneous multi-scale mapping framework remain unknown in both robotics and biology: in robotics because of the focus on single- or two-scale systems and limits in the scalability and open-field nature of current test environments and benchmark datasets; in biology because of technical limitations when recording from rodents during movement over large areas. New global spatial databases with visual information varying over several orders of magnitude in scale enable us to investigate this question for the first time in real-world environments. In particular, we investigate and answer the following questions: why have multi-scale representations, how many scales should there be, what should the size ratio between consecutive scales be and how does the absolute scale size affect performance? We answer these questions by developing and evaluating a homogeneous, multi-scale mapping framework mimicking aspects of the rodent multi-scale map, but using current robotic place recognition techniques at each scale. Results in large-scale real-world environments demonstrate multi-faceted and significant benefits for mapping and localization performance and identify the key factors that determine performance.

Place recognition from distant landmarks: human performance and maximum likelihood model

We present a simple behavioral experiment on human place recognition from a configuration of four visual landmarks. Participants were asked to navigate several paths, all involving a turn at one specific point, and while doing so incidentally learned the position of that turning point. In the test phase, they were asked to return to the turning point in a reduced environment leaving only the four landmarks visible. Results are compared to two versions of a maximum likelihood model of place recognition using either view-based or depth-based cues for place recognition. Only the depth-based model is in good qualitative agreement with the data. In particular, it reproduces landmark configuration-dependent effects of systematic bias and statistical error distribution as well as effects of approach direction. The model is based on a place code (depth and bearing of the landmarks at target location) and an egocentric working memory of surrounding space including current landmark position in a local, map-like representation. We argue that these elements are crucial for human place recognition.

Leveraging variable sensor spatial acuity with a homogeneous, multi-scale place recognition framework

Most robot navigation systems perform place recognition using a single-sensor modality and one, or at most two heterogeneous map scales. In contrast, mammals perform navigation by combining sensing from a wide variety of modalities including vision, auditory, olfactory and tactile senses with a multi-scale, homogeneous neural map of the environment. In this paper, we develop a multi-scale, multi-sensor system for mapping and place recognition that combines spatial localization hypotheses at different spatial scales from multiple different sensors to calculate an overall place recognition estimate. We evaluate the system's performance over three repeated 1.5-km day and night journeys across a university campus spanning outdoor and multi-level indoor environments, incorporating camera, WiFi and barometric sensory information. The system outperforms a conventional camera-only localization system, with the results demonstrating not only how combining multiple sensing modalities together improves performance, but also how combining these sensing modalities over multiple scales further improves performance over a single-scale approach. The multi-scale mapping framework enables us to analyze the naturally varying spatial acuity of different sensing modalities, revealing how the multi-scale approach captures each sensing modality at its optimal operation point where a single-scale approach does not, and enables us to then weight sensor contributions at different scales based on their utility for place recognition at that scale.

Schematic representations of local environmental space guide goal-directed navigation

To successfully navigate to a target, it is useful to be able to define its location at multiple levels of specificity. For example, the location of a favorite coffee mug can be described in terms of which room it is in, or in terms of where it is within the room. An appealing hypothesis is that these levels of description are retrieved from memory by accessing the same representation at progressively finer levels of granularity-first remembering the general location of an object and then "zooming in." Here we provide evidence for an alternative view, in which navigational behavior is guided by independent representations at multiple spatial scales. Subjects learned the locations of objects that were positioned within four visually distinct but geometrically similar buildings, which were in turn positioned within a broader virtual park. They were then tested on their knowledge of object location by asking them to navigate to the remembered location of each object. We examined errors during the test phase for confusions among geometrically analogous locations in different buildings-that is, navigating to the right location in the wrong building. We observed that subjects frequently made these confusions, which are analogous to remembering a passage's location on the page of a book but not remembering the page that the passage is on. This suggests that subjects were recalling the object's local location without recalling its global location. Further manipulations across seven experiments indicated that geometric confusions were observed even between buildings that were not metrically identical as long as geometrical equivalence could be defined. However, removing the walls so that the larger environment was no longer divided into subspaces abolished these errors. Taken together, our results suggest that human spatial memory contains two separable representations of "where" an object can be found: (i) a schematic map of where an object lies with respect to local landmarks and boundaries; (ii) a representation of the identity and location of each local environment.