Gridworld Pathfinding
Intermediate
Gridworld: agent navigates NxN grid from start to goal. Actions: [up, down, left, right]. Reward: -1 per step, +10 at goal. Walls block movement.
The problem, from first principles
Simple environment to test RL algorithms. Finite state/action spaces, deterministic dynamics, sparse rewards. Optimal policy finds shortest path.
From theory to code
Implement solve_gridworld(grid, start, goal, max_steps=1000) which:
- Uses value iteration or Q-learning
- grid: 2D array where 1=wall, 0=free
- Returns optimal policy as NxN array of actions
- Policy guides agent from any state to goal
Constraints
- 4 actions: up(0), down(1), left(2), right(3)
- Agent stays in place if action hits wall
- Deterministic transitions
- Goal is absorbing state
Hints
Hint 1: Value iteration
V(s) = max_a [R(s,a) + γ V(s')]Hint 2: Policy extraction
π(s) = argmax_a [R(s,a) + γ V(s')]Hint 3: Boundary handling
Out-of-bounds actions keep agent in placeKeywords
rlgridworldpathfindingmdp
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.