© DALL·E 2
Robotics and Biology Laboratory
Problem Solving using a Multi-Strategy Approach
Motivation
The selection of the appropriate cognitive or behavioral strategy from the mind's toolbox is a fundamental process that affects our ability to effectively solve tasks. Despite this, the mechanisms underlying strategy selection are not well understood. The rational meta-reasoning framework proposed by F. Lieder & T. L. Griffith, 2017, suggests that strategy selection is based on a subjective assessment of accuracy and costs, but these variables are often not directly accessible to the decision maker. Therefore, there is a need to better understand how these crucial parameters are cognitively and computationally implemented, in order to improve ecologically rational strategy selection. This research has the potential to enhance our understanding of cognitive processes and improve the ability of robotic systems to perform tasks more effectively.
Description of Work
This study aims to verify the concept of the "mind's toolbox" by simulating an environment where an agent must solve randomly generated mechanical puzzles (Lockboxes). The agent has a variety of strategies at their disposal, and must choose the best one for each Lockbox. The agent can change strategies mid-puzzle if needed.
Your tasks include:
- configuring the simulation
- creating the lockboxes
- identifying strategies that can solve the lockboxes
- implementing meta-strategies that select the most appropriate strategy for a given Lockbox
- benchmarking the different meta-strategies and evaluating the results
The scope is flexible and not limited to the listed tasks.
