多任务环境中的任务特定策略

本教程详细介绍了如何使用多任务策略和批量环境。

完成本教程后，你将能够编写使用不同权重集在不同设置中计算动作的策略。你还将能够并行执行不同的环境。

from tensordict import LazyStackedTensorDict
from tensordict.nn import TensorDictModule, TensorDictSequential
from torch import nn

from torchrl.envs import CatTensors, Compose, DoubleToFloat, ParallelEnv, TransformedEnv
from torchrl.envs.libs.dm_control import DMControlEnv
from torchrl.modules import MLP

我们设计了两个环境，一个仿人环境必须完成站立任务，另一个必须学习行走。

env1 = DMControlEnv("humanoid", "stand")
env1_obs_keys = list(env1.observation_spec.keys())
env1 = TransformedEnv(
    env1,
    Compose(
        CatTensors(env1_obs_keys, "observation_stand", del_keys=False),
        CatTensors(env1_obs_keys, "observation"),
        DoubleToFloat(
            in_keys=["observation_stand", "observation"],
            in_keys_inv=["action"],
        ),
    ),
)
env2 = DMControlEnv("humanoid", "walk")
env2_obs_keys = list(env2.observation_spec.keys())
env2 = TransformedEnv(
    env2,
    Compose(
        CatTensors(env2_obs_keys, "observation_walk", del_keys=False),
        CatTensors(env2_obs_keys, "observation"),
        DoubleToFloat(
            in_keys=["observation_walk", "observation"],
            in_keys_inv=["action"],
        ),
    ),
)

tdreset1 = env1.reset()
tdreset2 = env2.reset()

# With LazyStackedTensorDict, stacking is done in a lazy manner: the original tensordicts
# can still be recovered by indexing the main tensordict
tdreset = LazyStackedTensorDict.lazy_stack([tdreset1, tdreset2], 0)
assert tdreset[0] is tdreset1

print(tdreset[0])

策略

我们将设计一个策略，其中主干网络读取“observation”键。然后，如果存在的话，特定的子组件将读取堆叠 tensordicts 的“observation_stand”和“observation_walk”键，并将它们传递给专用的子网络。

action_dim = env1.action_spec.shape[-1]

policy_common = TensorDictModule(
    nn.Linear(67, 64), in_keys=["observation"], out_keys=["hidden"]
)
policy_stand = TensorDictModule(
    MLP(67 + 64, action_dim, depth=2),
    in_keys=["observation_stand", "hidden"],
    out_keys=["action"],
)
policy_walk = TensorDictModule(
    MLP(67 + 64, action_dim, depth=2),
    in_keys=["observation_walk", "hidden"],
    out_keys=["action"],
)
seq = TensorDictSequential(
    policy_common, policy_stand, policy_walk, partial_tolerant=True
)

我们来检查一下我们的序列是否为单个环境（站立）输出了动作。

seq(env1.reset())

我们来检查一下我们的序列是否为单个环境（行走）输出了动作。

seq(env2.reset())

这也适用于堆叠：现在 stand 和 walk 键已经消失了，因为它们并非所有 tensordicts 共享。但是 TensorDictSequential 仍然执行了操作。注意，主干网络是以向量化方式执行的——而不是在循环中——这更高效。

seq(tdreset)

并行执行不同任务

如果公共键值对共享相同的 specs（特别是它们的形状和 dtype 必须匹配：如果 observation 形状不同但指向同一个键，你将无法执行以下操作），我们可以并行化操作。

如果 ParallelEnv 接收单个环境创建函数，它将假定只需执行单个任务。如果提供函数列表，则将假定我们处于多任务设置。

def env1_maker():
    return TransformedEnv(
        DMControlEnv("humanoid", "stand"),
        Compose(
            CatTensors(env1_obs_keys, "observation_stand", del_keys=False),
            CatTensors(env1_obs_keys, "observation"),
            DoubleToFloat(
                in_keys=["observation_stand", "observation"],
                in_keys_inv=["action"],
            ),
        ),
    )


def env2_maker():
    return TransformedEnv(
        DMControlEnv("humanoid", "walk"),
        Compose(
            CatTensors(env2_obs_keys, "observation_walk", del_keys=False),
            CatTensors(env2_obs_keys, "observation"),
            DoubleToFloat(
                in_keys=["observation_walk", "observation"],
                in_keys_inv=["action"],
            ),
        ),
    )


env = ParallelEnv(2, [env1_maker, env2_maker])
assert not env._single_task

tdreset = env.reset()
print(tdreset)
print(tdreset[0])
print(tdreset[1])  # should be different

让我们将输出通过我们的网络。

tdreset = seq(tdreset)
print(tdreset)
print(tdreset[0])
print(tdreset[1])  # should be different but all have an "action" key


env.step(tdreset)  # computes actions and execute steps in parallel
print(tdreset)
print(tdreset[0])
print(tdreset[1])  # next_observation has now been written

Rollout

td_rollout = env.rollout(100, policy=seq, return_contiguous=False)

td_rollout[:, 0]  # tensordict of the first step: only the common keys are shown

td_rollout[0]  # tensordict of the first env: the stand obs is present

env.close()
del env