Agent Combines Egocentric And Allocentric Maps
Shani and Dayan publish on January 23, 2026 in PLoS Computational Biology a reinforcement-learning study that implements egocentric successor representations (SRs) composed with conventional allocentric SRs to support navigation in complex 2D environments. The paper shows that additive composition of egocentric and allocentric maps yields generalizable value functions, enabling faster learning and adaptation with fewer trials and reduced obstacle trapping. Data and code are available at the authors' GitHub repository.
Key Points
- 1Implements egocentric and allocentric successor representations in a single reinforcement-learning agent.
- 2Demonstrates additive composition yields generalized value functions enabling efficient policy learning across environments.
- 3Enables faster adaptation with fewer trials, reusable local rules, and reduced obstacle trapping.
Scoring Rationale
Strong peer-reviewed contribution combining egocentric and allocentric SRs, limited to simulated 2D navigation tasks and environments.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems