timelimit for reinforcement learning tasks

[YangyangFu]

Currently for dynamic system control tasks, we simply set the terminal condition as the maximum step length inside each episode. This may cause inefficient learning for on-policy algorithms like PPO.

Reference:

• Time Limit in Reinforcement Learning
• time limit wrapper in gym: https://github.com/openai/gym/blob/master/gym/wrappers/time_limit.py#L19

Good discussions:

• stablebaseline: [Bug] Infinite horizon tasks are handled like episodic tasks DLR-RM/stable-baselines3#284