Short Answer

Critiquing a Reward Function for Maze Navigation

An AI agent is being trained to navigate a maze. The reward function is defined as: +0.1 for each step taken, -100 for hitting a wall, and +1 for reaching the exit. The agent consistently learns to avoid walls but struggles to find the exit, often wandering aimlessly. Based on the principles of gradient estimation, identify the primary issue with this reward structure that contributes to the agent's poor performance and propose a specific numerical change to address it.

0

1

Updated 2025-10-10

Contributors are:

Who are from:

Tags

Ch.4 Alignment - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences

Evaluation in Bloom's Taxonomy

Cognitive Psychology

Psychology

Social Science

Empirical Science

Science