Lunar Lander Simulation
Advanced
Lunar Lander: soft-land spacecraft on moon. State: [x, y, vx, vy, angle, angular_vel]. Actions: [off, left-engine, main-engine, right-engine]. Reward: -1 per step, +200 landed, -100 crash.
The problem, from first principles
Continuous control with sparse rewards. Must learn to descend, manage fuel, stabilize. Solved when mean reward >= 200 over 100 episodes.
From theory to code
Implement simulate_lunar_lander(initial_state, actions, max_steps) which:
- Tracks fuel, position, velocity, attitude
- Applies thrust physics
- Computes rewards: step cost, landing bonus, crash penalty
- Returns trajectory with (state, reward) tuples
Constraints
- Gravity = 1.5 m/s^2
- Main engine thrust = 13.5 m/s^2, side = 0.75 m/s^2
- Fuel limited
- Must land soft: vx, vy < 0.1 m/s and angle < π/4
Hints
Hint 1: Thrust model
Acceleration = thrust_vector / mass - gravityHint 2: Landing condition
y < 0 and low velocity and stable angleHint 3: Fuel constraint
Track fuel consumption, episode ends if fuel depletedKeywords
rlcontinuous-controlphysics-simulationapplications
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.