Task Environment of an Agent

Its Pillars

Transclude of task-environment-of-an-agent-2026-04-03-18.07.50.excalidraw

Environment Types

1. Observability
   • Fully agent doesn’t need to guess anything about it, it can sense it fully.
   • Partial agent must guess, it has missing information.
2. State Transitions
   • Deterministic next state is fully determined only by the current action.
     (e.g., Chess)
   • Stochastic next state needs to take randomness into account. (e.g., Weather Prediction)
3. Previous States Effect
   • Episodic no need to later remember the current state, since upcoming states do not depend on it.
     (e.g., Spam Filter)
   • Sequential must save the current state, since upcoming states do depend on it.
     (e.g., Driving)
4. Environment Variability
   • Static the world does not change as the agent operates
     (e.g., Crossword Puzzle)
   • Dynamic the world does change as the agent operates
     (e.g., Driving)
5. States Frequency
   • Discrete their amount is limited, and thus are captured in discrete snapshots.
     (e.g., Chess)
   • Continuous their amount is infinite and must continuously be captured. (e.g., Robot arm movement)
6. Amount of Agents needed
   • Single only a single agent is needed.
     (e.g., solving Sodoku)
   • Multiagent multiple agents are required to interact together.
     (e.g., playing Soccer)

---

Connections