CartPole Control
Intermediate
CartPole: balance a pole on a moving cart. Classic benchmark for RL. Observations: [x, v, θ, ω]. Actions: [left, right]. Reward: +1 per step balanced.
The problem, from first principles
CartPole tests ability to learn stable control policies. Physics are simple but nonlinear. Solved when mean reward >= 195 over 100 episodes.
From theory to code
Implement simulate_cartpole(initial_state, actions, max_steps) which:
- Takes initial [x, v, θ, ω] and sequence of actions
- Simulates CartPole dynamics using Euler integration
- Returns trajectory of (state, reward) pairs
- Done when |x| > 2.4 or |θ| > π/12
Constraints
- Physics parameters from OpenAI Gym CartPole-v0
- Force = 10 N
- Gravity = 9.8 m/s^2
- Pole length = 0.5 m, mass = 0.1 kg
- Cart mass = 1.0 kg
Hints
Hint 1: State transition
Use equations of motion for inverted pendulum on cartHint 2: Done condition
Episode ends if x > 2.4 or θ > π/12Hint 3: Rewards
+1 per step while balanced, 0 when doneKeywords
rlcontrolsimulationapplications
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.