AttributeError: 'tuple' object has no attribute 'to'

[Yisikanadai]

/home/ubuntu417/anaconda3/envs/go2_rl/bin/python unitree_rl_gym-main/legged_gym/scripts/train.py --task=go2
Importing module 'gym_38' (/home/ubuntu417/unitree_go2/isaacgym/python/isaacgym/_bindings/linux-x86_64/gym_38.so)
Setting GYM_USD_PLUG_INFO_PATH to /home/ubuntu417/unitree_go2/isaacgym/python/isaacgym/_bindings/linux-x86_64/usd/plugInfo.json
PyTorch version 1.13.0+cu116
Using /home/ubuntu417/.cache/torch_extensions/py38_cu116 as PyTorch extensions root...
Emitting ninja build file /home/ubuntu417/.cache/torch_extensions/py38_cu116/gymtorch/build.ninja...
Building extension module gymtorch...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
Loading extension module gymtorch...
Device count 1
/home/ubuntu417/unitree_go2/isaacgym/python/isaacgym/_bindings/src/gymtorch
ninja: no work to do.
Setting seed: 1
Not connected to PVD
+++ Using GPU PhysX
Physics Engine: PhysX
Physics Device: cuda:0
GPU Pipeline: enabled
Traceback (most recent call last):
File "unitree_rl_gym-main/legged_gym/scripts/train.py", line 18, in
train(args)
File "unitree_rl_gym-main/legged_gym/scripts/train.py", line 14, in train
ppo_runner.learn(num_learning_iterations=train_cfg.runner.max_iterations, init_at_random_ep_len=True)
File "/home/ubuntu417/unitree_go2/rsl_rl-1.0.2/rsl_rl/runners/on_policy_runner.py", line 94, in learn
obs = obs.to(self.device)
AttributeError: 'tuple' object has no attribute 'to'

I print the obs

(tensor([ 0.3188, -0.7567, 0.4211, 0.2648, -0.1023, 0.1416, 0.0095, -0.0221,
-1.0391, 1.7729, -1.4516, -0.2500, 0.0288, -0.0544, -0.2970, -0.0498,
0.1930, -0.0523, 0.0227, 0.2263, -0.0120, 0.0081, -0.2732, 0.0702,
0.0331, -0.0871, 0.8401, 0.0096, -0.1943, 0.2382, -0.0160, -0.2962,
0.2006, 0.0689, 0.3692, -0.2438, 0.0000, 0.0000, 0.0000, 0.0000,
0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000],
device='cuda:0'), tensor([ 0.5073, 0.1631, -0.3483, -0.1181, -0.0402, 0.0271, -0.0341, -0.0391,
-0.9990, 0.4094, -1.3506, 0.1190, 0.0476, -0.1898, -0.2205, -0.0270,
0.1521, -0.2034, -0.0153, 0.2692, 0.3511, -0.0042, -0.0101, -0.0339,
-0.0519, 0.1259, 0.5441, -0.0084, -0.2618, 0.6837, 0.0273, -0.0903,
-0.6574, 0.0496, 0.0299, 0.1390, 0.0000, 0.0000, 0.0000, 0.0000,
0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000],
device='cuda:0'))

[craipy-hub]

Check the state of the network input and check the network type is LSTM or MLP

[Yisikanadai]

Check the state of the network input and check the network type is LSTM or MLP

Thanks to reply!!!

the network type is MLP
here is the network

Actor MLP: Sequential(
  (0): Linear(in_features=48, out_features=512, bias=True)
  (1): ELU(alpha=1.0)
  (2): Linear(in_features=512, out_features=256, bias=True)
  (3): ELU(alpha=1.0)
  (4): Linear(in_features=256, out_features=128, bias=True)
  (5): ELU(alpha=1.0)
  (6): Linear(in_features=128, out_features=12, bias=True)
)
Critic MLP: Sequential(
  (0): Linear(in_features=48, out_features=512, bias=True)
  (1): ELU(alpha=1.0)
  (2): Linear(in_features=512, out_features=256, bias=True)
  (3): ELU(alpha=1.0)
  (4): Linear(in_features=256, out_features=128, bias=True)
  (5): ELU(alpha=1.0)
  (6): Linear(in_features=128, out_features=1, bias=True)
)

here is the network input

obs[0]shape: torch.Size([48]) ,obs[1]shape: torch.Size([48])
critic_obs[0]shape: torch.Size([48]) ,critic_obs[1]shape: torch.Size([48])

the learn code is here

    def learn(self, num_learning_iterations, init_at_random_ep_len=False):
        # initialize writer
        if self.log_dir is not None and self.writer is None:
            self.writer = SummaryWriter(log_dir=self.log_dir, flush_secs=10)
        if init_at_random_ep_len:
            self.env.episode_length_buf = torch.randint_like(self.env.episode_length_buf, high=int(self.env.max_episode_length))
        obs = self.env.get_observations()
        privileged_obs = self.env.get_privileged_observations()
        critic_obs = privileged_obs if privileged_obs is not None else obs
        print("obs[0]shape:",obs[0].shape,",obs[1]shape:",obs[1].shape)
        print("critic_obs[0]shape:",obs[0].shape,",critic_obs[1]shape:",obs[1].shape)
        obs = obs[0].to(self.device)      ## change
        critic_obs = critic_obs[0].to(self.device)         ## change
        self.alg.actor_critic.train() # switch to train mode (for dropout for example)

        ep_infos = []
        rewbuffer = deque(maxlen=100)
        lenbuffer = deque(maxlen=100)
        cur_reward_sum = torch.zeros(self.env.num_envs, dtype=torch.float, device=self.device)
        cur_episode_length = torch.zeros(self.env.num_envs, dtype=torch.float, device=self.device)

        tot_iter = self.current_learning_iteration + num_learning_iterations
        for it in range(self.current_learning_iteration, tot_iter):
            start = time.time()
            # Rollout
            with torch.inference_mode():
                for i in range(self.num_steps_per_env):
                    actions = self.alg.act(obs, critic_obs)
                    obs, privileged_obs, rewards, dones, infos = self.env.step(actions)
                    critic_obs = privileged_obs if privileged_obs is not None else obs
                    obs, critic_obs, rewards, dones = obs.to(self.device), critic_obs.to(self.device), rewards.to(self.device), dones.to(self.device)
                    self.alg.process_env_step(rewards, dones, infos)
.......

I try to change obs = obs.to(self.device) to obs = obs[0].to(self.device)
but get a new error

Actor MLP: Sequential(
  (0): Linear(in_features=48, out_features=512, bias=True)
  (1): ELU(alpha=1.0)
  (2): Linear(in_features=512, out_features=256, bias=True)
  (3): ELU(alpha=1.0)
  (4): Linear(in_features=256, out_features=128, bias=True)
  (5): ELU(alpha=1.0)
  (6): Linear(in_features=128, out_features=12, bias=True)
)
Critic MLP: Sequential(
  (0): Linear(in_features=48, out_features=512, bias=True)
  (1): ELU(alpha=1.0)
  (2): Linear(in_features=512, out_features=256, bias=True)
  (3): ELU(alpha=1.0)
  (4): Linear(in_features=256, out_features=128, bias=True)
  (5): ELU(alpha=1.0)
  (6): Linear(in_features=128, out_features=1, bias=True)
)
obs[0]shape: torch.Size([48]) ,obs[1]shape: torch.Size([48])
critic_obs[0]shape: torch.Size([48]) ,critic_obs[1]shape: torch.Size([48])
Traceback (most recent call last):
  File "legged_gym/scripts/train.py", line 18, in <module>
    train(args)
  File "legged_gym/scripts/train.py", line 14, in train
    ppo_runner.learn(num_learning_iterations=train_cfg.runner.max_iterations, init_at_random_ep_len=True)
  File "/home/ubuntu417/unitree_go2/rsl_rl-1.0.2/rsl_rl/runners/on_policy_runner.py", line 114, in learn
    obs, privileged_obs, rewards, dones, infos = self.env.step(actions)
  File "/home/ubuntu417/unitree_go2/unitree_rl_gym-main/legged_gym/envs/base/legged_robot.py", line 74, in step
    self.post_physics_step()
  File "/home/ubuntu417/unitree_go2/unitree_rl_gym-main/legged_gym/envs/base/legged_robot.py", line 110, in post_physics_step
    self.compute_observations() # in some cases a simulation step might be required to refresh some obs (for example body positions)
  File "/home/ubuntu417/unitree_go2/unitree_rl_gym-main/legged_gym/envs/base/legged_robot.py", line 182, in compute_observations
    self.obs_buf = torch.cat((  self.base_lin_vel * self.obs_scales.lin_vel,
RuntimeError: Tensors must have same number of dimensions: got 2 and 1

I find the code is here

    def compute_observations(self):
        """ Computes observations
        """
        print((self.base_lin_vel * self.obs_scales.lin_vel).shape)
        print((self.base_ang_vel  * self.obs_scales.ang_vel).shape)
        print((self.commands[:, :3] * self.commands_scale).shape)
        print(((self.dof_pos - self.default_dof_pos) * self.obs_scales.dof_pos).shape)
        print((self.dof_vel * self.obs_scales.dof_vel).shape)
        print((self.actions).shape)
        self.obs_buf = torch.cat((  self.base_lin_vel * self.obs_scales.lin_vel,
                                    self.base_ang_vel  * self.obs_scales.ang_vel,
                                    self.projected_gravity,
                                    self.commands[:, :3] * self.commands_scale,
                                    (self.dof_pos - self.default_dof_pos) * self.obs_scales.dof_pos,
                                    self.dof_vel * self.obs_scales.dof_vel,
                                    self.actions
                                    ),dim=-1)
        # add perceptive inputs if not blind
        # add noise if needed
        if self.add_noise:
            self.obs_buf += (2 * torch.rand_like(self.obs_buf) - 1) * self.noise_scale_vec

the output is

torch.Size([4096, 3])
torch.Size([4096, 3])
torch.Size([4096, 3])
torch.Size([4096, 12])
torch.Size([4096, 12])
torch.Size([12])