liveBook · Manning

Partially observable environment

As explained in Sutskever's List

An environment in which an agent observes only a limited or incomplete representation of the underlying state.

A partially observable environment reveals only part of its underlying situation. In Mini-Pacman, the agent sees a limited viewport, so pellets, walls, and ghosts disappear from view and must be tracked through memory. section 7.4.1

Mini-Pacman

The agent navigates a maze, collects pellets, and avoids moving ghosts. Because unseen objects may move while hidden, success requires integrating observations over time. An RMC agent scored 677 against an LSTM agent's 550 in the limited-view setting, indicating a benefit from relational memory. section 7.4.1

How Sutskever's List covers it

How deep it goes
Defines partial observability through a limited viewport and demonstrates its memory demands, without formal state-estimation treatment.
Examples
Uses Mini-Pacman, hidden maze objects, moving ghosts, and limited-view scores to illustrate the problem.
What it assumes
Assumes basic familiarity with agents, rewards, penalties, memory, and reinforcement learning.
What you can do after
Evaluate how memory architectures handle hidden objects and changing relationships across sequential observations.
How it connects
Connects partial observability with reinforcement learning, recurrent memory, LSTMs, and relational reasoning.
Cover of Sutskever's List
liveBookmanning.com
Read Sutskever's List on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Sutskever's List. The section links point to the parts of the book it draws on.