Backward error in ddpg algorithm

reinforcement-learning
2

You might be trying to calculate the gradients using stale forward activations with already updated parameters as described in this post.