5DV248 · Trustworthy Human-AI Interaction · Laboratory

Mountain Rescue: A Goal-Based Agent

A rescue robot must reach a stranded hiker and bring them safely back to base. It has to reason about the state of the world, represent the goal, and find a sequence of actions that achieves it.

The mission

A hiker is stranded on a mountain trail. A rescue robot starts at the base station. Rocks and collapsed paths block some routes. Your task is to observe how a goal-based agent plans a route, performs the rescue, and returns to base.

Goal state: the robot is back at the base and the hiker has been rescued. Reaching the hiker is only an intermediate state, not mission completion.

What makes it goal-based?

The agent does not have a fixed rule such as “always move north” or “go to the nearest open cell”. It considers possible future states and selects actions that lead to its goal.

Current state Goal description Search / planning

This follows the architecture introduced in Lecture 1, slides 43–44: state and goal information guide the choice of an action sequence.

Rescue environment

The robot is at the base. The hiker is at the rescue location. Dark cells are impassable rocks or collapsed paths.

Turn editing on, then click cells to add/remove obstacles.
Base Hiker Obstacle Search frontier Explored Planned route
0Actions executed
0States explored
—Plan actions left
Ready. Select Find a plan to let the agent search for an action sequence.

Agent's plan

The search considers both the robot's position and whether the hiker has been rescued.

No plan generated yet.
Current positionBase (8, 0)
Hiker rescued?No
Goal conditionAt base AND rescued
Planning methodBreadth-first search

Investigate the agent

  1. Before clicking Find a plan, predict what the robot must do to achieve the goal.
  2. Inspect the explored cells and the green route. Why does the robot need to visit the hiker before returning to base?
  3. Press Step repeatedly. Identify the action that changes the state from “not rescued” to “rescued”.
  4. After the rescue, explain why the robot continues moving. Has it achieved the goal yet?
  5. Turn on obstacle editing, change the terrain, and plan again. How does the action sequence change while the goal remains the same?

Connect to the lecture

Slide 43 — Goal information. A goal describes a desirable situation. Here, “the hiker is rescued and the robot is at base” specifies what counts as success.

Slide 44 — Goal-based agent function. The agent keeps track of the world state and its goal, then chooses an action that will eventually lead to goal achievement. Search and planning find the action sequence.

Why is this not just a reflex agent?

A reflex agent selects an action from the current percept using condition–action rules. This agent instead searches over possible future states. The next action is selected as part of a plan directed toward a stated goal.

Why is this not yet a utility-based agent?

The mission has a goal condition: either the hiker is rescued and the robot is back at base, or the goal is not achieved. This laboratory does not assign different numerical values to outcomes or trade off competing objectives. That distinction leads to utility-based agents in the next part of the lecture.