feat: remove the `use_history` field, add `observations_history` to support history observations, and adjust the related logic to handle the case of empty history observations

src/rl_sar/library/rl_sdk/rl_sdk.cpp

@@ -421,11 +421,18 @@ void RL::ReadYaml(std::string robot_name)
     this->params.framework = config["framework"].as<std::string>();
     int rows = config["rows"].as<int>();
     int cols = config["cols"].as<int>();
-    this->params.use_history = config["use_history"].as<bool>();
     this->params.dt = config["dt"].as<double>();
     this->params.decimation = config["decimation"].as<int>();
     this->params.num_observations = config["num_observations"].as<int>();
     this->params.observations = ReadVectorFromYaml<std::string>(config["observations"]);
+    if (config["observations_history"].IsNull())
+    {
+        this->params.observations_history = {};
+    }
+    else
+    {
+        this->params.observations_history = ReadVectorFromYaml<int>(config["observations_history"]);
+    }
     this->params.clip_obs = config["clip_obs"].as<double>();
     if (config["clip_actions_lower"].IsNull() && config["clip_actions_upper"].IsNull())
     {

src/rl_sar/library/rl_sdk/rl_sdk.hpp

@@ -72,11 +72,11 @@ struct ModelParams
 {
     std::string model_name;
     std::string framework;
-    bool use_history;
     double dt;
     int decimation;
     int num_observations;
     std::vector<std::string> observations;
+    std::vector<int> observations_history;
     double damping;
     double stiffness;
     double action_scale;

src/rl_sar/models/a1_isaacgym/config.yaml

@@ -3,11 +3,11 @@ a1_isaacgym:
   framework: "isaacgym"
   rows: 4
   cols: 3
-  use_history: True
   dt: 0.005
   decimation: 4
   num_observations: 45
   observations: ["ang_vel", "gravity_vec", "commands", "dof_pos", "dof_vel", "actions"]
+  observations_history: [0, 1, 2, 3, 4, 5]
   clip_obs: 100.0
   clip_actions_lower: [-100, -100, -100,
                        -100, -100, -100,

src/rl_sar/models/a1_isaacsim/config.yaml

@@ -3,11 +3,11 @@ a1_isaacsim:
   framework: "isaacsim"
   rows: 4
   cols: 3
-  use_history: False
   dt: 0.005
   decimation: 4
   num_observations: 45
   observations: ["ang_vel", "gravity_vec", "commands", "dof_pos", "dof_vel", "actions"]
+  observations_history: []
   clip_obs: 100.0
   clip_actions_lower: [-100, -100, -100,
                        -100, -100, -100,

src/rl_sar/models/go2_isaacgym/config.yaml

@@ -3,11 +3,11 @@ go2_isaacgym:
   framework: "isaacgym"
   rows: 4
   cols: 3
-  use_history: True
   dt: 0.005
   decimation: 4
   num_observations: 45
   observations: ["commands", "ang_vel", "gravity_vec", "dof_pos", "dof_vel", "actions"]
+  observations_history: [5, 4, 3, 2, 1, 0]
   clip_obs: 100.0
   clip_actions_lower: [-100, -100, -100,
                        -100, -100, -100,

src/rl_sar/models/gr1t1_isaacgym/config.yaml

@@ -3,11 +3,11 @@ gr1t1_isaacgym:
   framework: "isaacgym"
   rows: 2
   cols: 5
-  use_history: False
   dt: 0.001
   decimation: 20
   num_observations: 39
   observations: ["ang_vel", "gravity_vec", "commands", "dof_pos", "dof_vel", "actions"]
+  observations_history: []
   clip_obs: 100.0
   clip_actions_lower: [-0.4391, -1.0491, -2.0991, -0.4391, -1.3991,
                        -1.1391, -1.0491, -2.0991, -0.4391, -1.3991]

src/rl_sar/models/gr1t1_isaacsim/config.yaml

@@ -3,11 +3,11 @@ gr1t1_isaacsim:
   framework: "isaacsim"
   rows: 2
   cols: 5
-  use_history: False
   dt: 0.001
   decimation: 20
   num_observations: 39
   observations: ["ang_vel", "gravity_vec", "commands", "dof_pos", "dof_vel", "actions"]
+  observations_history: []
   clip_obs: 100.0
   clip_actions_lower: [-0.4391, -1.0491, -2.0991, -0.4391, -1.3991,
                        -1.1391, -1.0491, -2.0991, -0.4391, -1.3991]

src/rl_sar/models/gr1t2_isaacgym/config.yaml

@@ -3,11 +3,11 @@ gr1t2_isaacgym:
   framework: "isaacgym"
   rows: 2
   cols: 5
-  use_history: False
   dt: 0.001
   decimation: 20
   num_observations: 39
   observations: ["ang_vel", "gravity_vec", "commands", "dof_pos", "dof_vel", "actions"]
+  observations_history: []
   clip_obs: 100.0
   clip_actions_lower: [-0.4391, -1.0491, -2.0991, -0.4391, -1.3991,
                        -1.1391, -1.0491, -2.0991, -0.4391, -1.3991]

src/rl_sar/models/gr1t2_isaacsim/config.yaml

@@ -3,11 +3,11 @@ gr1t2_isaacsim:
   framework: "isaacsim"
   rows: 2
   cols: 5
-  use_history: False
   dt: 0.001
   decimation: 20
   num_observations: 39
   observations: ["ang_vel", "gravity_vec", "commands", "dof_pos", "dof_vel", "actions"]
+  observations_history: []
   clip_obs: 100.0
   clip_actions_lower: [-0.4391, -1.0491, -2.0991, -0.4391, -1.3991,
                        -1.1391, -1.0491, -2.0991, -0.4391, -1.3991]

src/rl_sar/scripts/rl_sdk.py

@@ -64,11 +64,11 @@ class ModelParams:
     def __init__(self):
         self.model_name = None
         self.framework = None
-        self.use_history = None
         self.dt = None
         self.decimation = None
         self.num_observations = None
         self.observations = None
+        self.observations_history = None
         self.damping = None
         self.stiffness = None
         self.action_scale = None
@@ -378,16 +378,20 @@ class RL:
         self.params.framework = config["framework"]
         rows = config["rows"]
         cols = config["cols"]
-        self.params.use_history = config["use_history"]
         self.params.dt = config["dt"]
         self.params.decimation = config["decimation"]
         self.params.num_observations = config["num_observations"]
         self.params.observations = config["observations"]
+        if config["observations_history"] is None:
+            self.params.observations_history = None
+        else:
+            self.params.observations_history = config["observations_history"]
+
         self.params.clip_obs = config["clip_obs"]
         self.params.action_scale = config["action_scale"]
         self.params.hip_scale_reduction = config["hip_scale_reduction"]
         self.params.hip_scale_reduction_indices = config["hip_scale_reduction_indices"]
-        if config["clip_actions_upper"] is None and config["clip_actions_upper"] is None:
+        if config["clip_actions_lower"] is None and config["clip_actions_upper"] is None:
             self.params.clip_actions_upper = None
             self.params.clip_actions_lower = None
         else:

src/rl_sar/scripts/rl_sim.py

@@ -36,8 +36,8 @@ class RL_Sim(RL):
         self.ReadYaml(self.robot_name)
 
         # history
-        if self.params.use_history:
+        if self.params.observations_history is None:
-            self.history_obs_buf = ObservationBuffer(1, self.params.num_observations, 6)
+            self.history_obs_buf = ObservationBuffer(1, self.params.num_observations, len(self.params.observations_history))
 
         # Due to the fact that the robot_state_publisher sorts the joint names alphabetically,
         # the mapping table is established according to the order defined in the YAML file
@@ -202,9 +202,9 @@ class RL_Sim(RL):
     def Forward(self):
         torch.set_grad_enabled(False)
         clamped_obs = self.ComputeObservation()
-        if self.params.use_history:
+        if self.params.observations_history is None:
             self.history_obs_buf.insert(clamped_obs)
-            history_obs = self.history_obs_buf.get_obs_vec(np.arange(6))
+            history_obs = self.history_obs_buf.get_obs_vec(self.params.observations_history)
             actions = self.model.forward(history_obs)
         else:
             actions = self.model.forward(clamped_obs)

src/rl_sar/src/rl_real_a1.cpp

@@ -12,7 +12,10 @@ RL_Real::RL_Real() : unitree_safe(UNITREE_LEGGED_SDK::LeggedType::A1), unitree_u
     this->ReadYaml(this->robot_name);
 
     // history
-    this->history_obs_buf = ObservationBuffer(1, this->params.num_observations, 6);
+    if (!this->params.observations_history.empty())
+    {
+        this->history_obs_buf = ObservationBuffer(1, this->params.num_observations, this->params.observations_history.size());
+    }
 
     this->unitree_udp.InitCmdData(this->unitree_low_command);
 
@@ -176,10 +179,17 @@ torch::Tensor RL_Real::Forward()
 
     torch::Tensor clamped_obs = this->ComputeObservation();
 
+    torch::Tensor actions;
+    if (!this->params.observations_history.empty())
+    {
         this->history_obs_buf.insert(clamped_obs);
-    this->history_obs = this->history_obs_buf.get_obs_vec({0, 1, 2, 3, 4, 5});
+        this->history_obs = this->history_obs_buf.get_obs_vec(this->params.observations_history);
-
+        actions = this->model.forward({this->history_obs}).toTensor();
-    torch::Tensor actions = this->model.forward({this->history_obs}).toTensor();
+    }
+    else
+    {
+        actions = this->model.forward({clamped_obs}).toTensor();
+    }
 
     if (this->params.clip_actions_upper.numel() != 0 && this->params.clip_actions_lower.numel() != 0)
     {

src/rl_sar/src/rl_real_go2.cpp

@@ -12,7 +12,10 @@ void RL_Real::RL_Real()
     this->ReadYaml(this->robot_name);
 
     // history
-    this->history_obs_buf = ObservationBuffer(1, this->params.num_observations, 6);
+    if (!this->params.observations_history.empty())
+    {
+        this->history_obs_buf = ObservationBuffer(1, this->params.num_observations, this->params.observations_history.size());
+    }
 
     // init robot
     this->InitRobotStateClient();
@@ -187,10 +190,17 @@ torch::Tensor RL_Real::Forward()
 
     torch::Tensor clamped_obs = this->ComputeObservation();
 
+    torch::Tensor actions;
+    if (!this->params.observations_history.empty())
+    {
         this->history_obs_buf.insert(clamped_obs);
-    this->history_obs = this->history_obs_buf.get_obs_vec({5, 4, 3, 2, 1, 0});
+        this->history_obs = this->history_obs_buf.get_obs_vec(this->params.observations_history);
-
+        actions = this->model.forward({this->history_obs}).toTensor();
-    torch::Tensor actions = this->model.forward({this->history_obs}).toTensor();
+    }
+    else
+    {
+        actions = this->model.forward({clamped_obs}).toTensor();
+    }
 
     if (this->params.clip_actions_upper.numel() != 0 && this->params.clip_actions_lower.numel() != 0)
     {

src/rl_sar/src/rl_sim.cpp

@@ -12,9 +12,9 @@ RL_Sim::RL_Sim()
     this->ReadYaml(this->robot_name);
 
     // history
-    if (this->params.use_history)
+    if (!this->params.observations_history.empty())
     {
-        this->history_obs_buf = ObservationBuffer(1, this->params.num_observations, 6);
+        this->history_obs_buf = ObservationBuffer(1, this->params.num_observations, this->params.observations_history.size());
     }
 
     // Due to the fact that the robot_state_publisher sorts the joint names alphabetically,
@@ -247,13 +247,14 @@ void RL_Sim::RunModel()
 torch::Tensor RL_Sim::Forward()
 {
     torch::autograd::GradMode::set_enabled(false);
+
     torch::Tensor clamped_obs = this->ComputeObservation();
+
     torch::Tensor actions;
-    if (this->params.use_history)
+    if (!this->params.observations_history.empty())
     {
         this->history_obs_buf.insert(clamped_obs);
-        // TODO-devel-go2 这里要找一种方法适配不同的顺序，不能直接改这里，会导致a1的模型不可用
+        this->history_obs = this->history_obs_buf.get_obs_vec(this->params.observations_history);
-        this->history_obs = this->history_obs_buf.get_obs_vec({5, 4, 3, 2, 1, 0});
         actions = this->model.forward({this->history_obs}).toTensor();
     }
     else