The detour problem in a stochastic environment: Tolman revisited

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Behavioral Economics

Background:

• Human planning and replanning are crucial for navigating dynamic and uncertain environments.
• Understanding how individuals learn, represent, and adapt to environmental changes is key to cognitive modeling.

Purpose of the Study:

• To investigate human planning and replanning behaviors in an unknown stochastic grid world.
• To compare various computational models explaining learning, representation, and planning strategies.
• To determine the best model for describing behavioral data, especially during replanning.

Main Methods:

• Development of a grid world task with unknown stochastic properties and dynamic path alterations.
• Participant navigation from random start to goal positions, optimizing reward acquisition.
• Comparison of 12 distinct computational models, including heuristic and reinforcement learning approaches.
• Model comparison using behavioral data from planning and replanning trials.

Main Results:

• Majority of participants demonstrated optimal planning capabilities.
• Individuals successfully revised their plans when encountering unexpected path blockages.
• Model-based reinforcement learning models significantly outperformed heuristic models in explaining replanning behavior.

Conclusions:

• Humans exhibit robust planning and adaptive replanning skills in novel, uncertain environments.
• Model-based reinforcement learning provides a superior framework for understanding human decision-making under uncertainty and change.
• The study highlights the importance of learning and adaptation in human planning.