# Composing egocentric and allocentric maps for flexible navigation

## TL;DR

This paper shows how combining self-centered and world-centered maps improves navigation efficiency and adaptability in changing environments.

## Contribution

The paper introduces a reinforcement learning agent that combines egocentric and allocentric successor representations for flexible navigation.

## Key findings

- The agent learns generalizable value functions from both egocentric and allocentric maps.
- Combining the maps allows faster adaptation to new environments and better obstacle avoidance.
- Egocentric representations capture reusable local rules, while allocentric ones handle global direction.

## Abstract

Egocentric representations of the environment have historically been relegated to being used only for simple forms of spatial behaviour such as stimulus-response learning. However, in the many cases that critical aspects of policies are best defined relative to the self, egocentric representations can be advantageous. Furthermore, there is evidence that forms of egocentric representation might exist in the wider hippocampal formation. Nevertheless, egocentric representations have yet to be fully incorporated as a component of modern navigational methods. Here we investigate egocentric successor representations (SRs) and their combination with allocentric representations. We build a reinforcement learning agent that combines an egocentric SR with a conventional allocentric SR to navigate complex 2D environments. We demonstrate that the agent learns generalisable egocentric and allocentric value functions which, even when only additively composed, allow it to learn policies efficiently and to adapt to new environments quickly. Our work shows the benefit for egocentric relational structure to be captured, as well as allocentric. We offer a new perspective on how cognitive maps could usefully be composed from multiple simple maps representing associations between state features defined in different reference frames.

Humans and animals use two kinds of representation to navigate. One is world-centered or allocentric: a bird’s-eye view of objects and places within which the subject is positioned. The other is self-centered or egocentric: with the locations of objects and places being referenced to the subject themselves, e.g., “the wall on my right.” Most neuroscientific research into navigation leans heavily on allocentric representations, likening them to “maps” that represent associations between objects and places and can facilitate complex planning. Meanwhile egocentric representations are viewed as being rather simple, and mainly employed for stimulus-response learning. Here, we show the benefits of constructing and using a second, egocentric, map alongside the conventional, allocentric, one. We designed a simple learning agent that builds both sorts of map, and uses them cooperatively. The egocentric map captures local rules that repeat across different layouts of environments, while the allocentric map handles overall direction. Since the local rules are reusable, the combined agent adapts faster than a purely allocentric one when environments change: it needs fewer trials to find good paths and avoids getting “trapped” by obstacles. Our approach highlights the benefit for planning of using multiple maps in different reference frames.