Proximal Policy Optimization

OpenAI released Proximal Policy Optimization (PPO), a simpler reinforcement learning method with a clipped objective that limits policy updates.

PPO removes the need for adaptive KL penalties and is easier to implement and tune than TRPO and ACER.

OpenAI reports strong results in continuous control and near-competitive performance on Atari tasks.

PPO became OpenAI’s default RL algorithm and powers interactive robot control demos.