DDPG: The Critic Target
Beginner
The problem, from first principles
DDPG learns a deterministic actor and a critic that scores actions. The critic is trained toward a one-step target that uses the target actor's action and the target critic's value at the next state.
From theory to code
Implement ddpg_critic_target(r, gamma, done, q_next), the one-step critic target.
Constraints
- No bootstrap after a terminal state.
Hints
Hint 1
Compute r + gamma * (1 - done) * q_next.
Keywords
research-papersreinforcement-learningcontinuous-controlactor-critic
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.