# Omniverse Isaac Sim tutorial: Creating New RL Environment
# https://docs.omniverse.nvidia.com/isaacsim/latest/tutorial_gym_new_rl_example.html

# instantiate the VecEnvBase and create the task
from omni.isaac.gym.vec_env import VecEnvBase  # isort: skip
env = VecEnvBase(headless=True)

from cartpole_task import CartpoleTask  # isort: skip
task = CartpoleTask(name="Cartpole")
env.set_task(task, backend="torch")


import torch
import torch.nn as nn

# import the skrl components to build the RL system
from skrl.agents.torch.ppo import PPO, PPO_DEFAULT_CONFIG
from skrl.envs.wrappers.torch import wrap_env
from skrl.memories.torch import RandomMemory
from skrl.models.torch import DeterministicMixin, GaussianMixin, Model
from skrl.trainers.torch import SequentialTrainer
from skrl.utils import set_seed


# seed for reproducibility
set_seed()  # e.g. `set_seed(42)` for fixed seed


# define models (stochastic and deterministic models) using mixins
class Policy(GaussianMixin, Model):
    def __init__(self, observation_space, action_space, device, clip_actions=False,
                 clip_log_std=True, min_log_std=-20, max_log_std=2, reduction="sum"):
        Model.__init__(self, observation_space, action_space, device)
        GaussianMixin.__init__(self, clip_actions, clip_log_std, min_log_std, max_log_std, reduction)

        self.net = nn.Sequential(nn.Linear(self.num_observations, 64),
                                 nn.Tanh(),
                                 nn.Linear(64, 64),
                                 nn.Tanh(),
                                 nn.Linear(64, self.num_actions))
        self.log_std_parameter = nn.Parameter(torch.zeros(self.num_actions))

    def compute(self, inputs, role):
        return torch.tanh(self.net(inputs["states"])), self.log_std_parameter, {}

class Value(DeterministicMixin, Model):
    def __init__(self, observation_space, action_space, device, clip_actions=False):
        Model.__init__(self, observation_space, action_space, device)
        DeterministicMixin.__init__(self, clip_actions)

        self.net = nn.Sequential(nn.Linear(self.num_observations, 64),
                                 nn.Tanh(),
                                 nn.Linear(64, 64),
                                 nn.Tanh(),
                                 nn.Linear(64, 1))

    def compute(self, inputs, role):
        return self.net(inputs["states"]), {}


# load and wrap the environment
env = wrap_env(env)

device = env.device


# instantiate a memory as rollout buffer (any memory can be used for this)
memory = RandomMemory(memory_size=1000, num_envs=env.num_envs, device=device)


# instantiate the agent's models (function approximators).
# PPO requires 2 models, visit its documentation for more details
# https://skrl.readthedocs.io/en/latest/api/agents/ppo.html#models
models = {}
models["policy"] = Policy(env.observation_space, env.action_space, device, clip_actions=True)
models["value"] = Value(env.observation_space, env.action_space, device)

# initialize models' parameters (weights and biases)
for model in models.values():
    model.init_parameters(method_name="normal_", mean=0.0, std=0.1)


# configure and instantiate the agent (visit its documentation to see all the options)
# https://skrl.readthedocs.io/en/latest/api/agents/ppo.html#configuration-and-hyperparameters
cfg = PPO_DEFAULT_CONFIG.copy()
cfg["rollouts"] = 1000  # memory_size
cfg["learning_epochs"] = 20
cfg["mini_batches"] = 1
cfg["discount_factor"] = 0.99
cfg["lambda"] = 0.95
cfg["learning_rate"] = 1e-3
cfg["grad_norm_clip"] = 1.0
cfg["ratio_clip"] = 0.2
cfg["value_clip"] = 0.2
cfg["clip_predicted_values"] = False
cfg["entropy_loss_scale"] = 0.0
cfg["value_loss_scale"] = 0.5
cfg["kl_threshold"] = 0
# logging to TensorBoard and write checkpoints (in timesteps)
cfg["experiment"]["write_interval"] = 1000
cfg["experiment"]["checkpoint_interval"] = 10000
cfg["experiment"]["directory"] = "runs/torch/Cartpole"

agent = PPO(models=models,
            memory=memory,
            cfg=cfg,
            observation_space=env.observation_space,
            action_space=env.action_space,
            device=device)


# configure and instantiate the RL trainer
cfg_trainer = {"timesteps": 100000, "headless": True}
trainer = SequentialTrainer(cfg=cfg_trainer, env=env, agents=agent)

# start training
trainer.train()