Merge pull request #60 from MuammerBay/feat/reach-task-new-asset

Adds reach task for new asset

Author: louislelay

source/SO_100/SO_100/tasks/lift/joint_pos_env_cfg.py

@@ -25,6 +25,10 @@
 
 from isaaclab.markers.config import FRAME_MARKER_CFG  # isort: skip
 
+# ----------------------------------------------------------------
+# --------------- LycheeAI live asset ----------------------------
+# ----------------------------------------------------------------
+
 
 @configclass
 class SoArm100LiftCubeEnvCfg(LiftEnvCfg):

source/SO_100/SO_100/tasks/reach/__init__.py

@@ -16,7 +16,10 @@
 # Register Gym environments.
 ##
 
-# Register the SO-100 Reach environment
+# ----------------------------------------------------------------
+# --------------- LycheeAI live asset ----------------------------
+# ----------------------------------------------------------------
+
 gym.register(
     id="SO-ARM100-Reach-v0",
     entry_point="isaaclab.envs:ManagerBasedRLEnv",
@@ -38,3 +41,29 @@
     },
     disable_env_checker=True,
 )
+
+# ----------------------------------------------------------------
+# --------------- ROSCON ES 2025 asset ---------------------------
+# ----------------------------------------------------------------
+
+# Joint position controller
+
+gym.register(
+    id="SO-ARM100-Reach-ROSCON-v0",
+    entry_point="isaaclab.envs:ManagerBasedRLEnv",
+    kwargs={
+        "env_cfg_entry_point": f"{__name__}.joint_pos_env_cfg:SoArm100ReachRosConEnvCfg",
+        "rsl_rl_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:ReachRosConPPORunnerCfg",
+    },
+    disable_env_checker=True,
+)
+
+gym.register(
+    id="SO-ARM100-Reach-ROSCON-Play-v0",
+    entry_point="isaaclab.envs:ManagerBasedRLEnv",
+    kwargs={
+        "env_cfg_entry_point": f"{__name__}.joint_pos_env_cfg:SoArm100ReachRosConEnvCfg_PLAY",
+        "rsl_rl_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:ReachRosConPPORunnerCfg",
+    },
+    disable_env_checker=True,
+)

source/SO_100/SO_100/tasks/reach/agents/rsl_rl_ppo_cfg.py

@@ -45,3 +45,34 @@ class ReachPPORunnerCfg(RslRlOnPolicyRunnerCfg):
         desired_kl=0.01,
         max_grad_norm=1.0,
     )
+
+
+@configclass
+class ReachRosConPPORunnerCfg(RslRlOnPolicyRunnerCfg):
+    num_steps_per_env = 24
+    max_iterations = 1000
+    save_interval = 50
+    experiment_name = "so_arm100_reach"
+    run_name = ""
+    resume = False
+    empirical_normalization = False
+    policy = RslRlPpoActorCriticCfg(
+        init_noise_std=1.0,
+        actor_hidden_dims=[64, 64],
+        critic_hidden_dims=[64, 64],
+        activation="elu",
+    )
+    algorithm = RslRlPpoAlgorithmCfg(
+        value_loss_coef=1.0,
+        use_clipped_value_loss=True,
+        clip_param=0.2,
+        entropy_coef=0.001,
+        num_learning_epochs=8,
+        num_mini_batches=4,
+        learning_rate=1.0e-3,
+        schedule="adaptive",
+        gamma=0.99,
+        lam=0.95,
+        desired_kl=0.01,
+        max_grad_norm=1.0,
+    )

source/SO_100/SO_100/tasks/reach/joint_pos_env_cfg.py

@@ -24,6 +24,10 @@
 # Scene definition
 ##
 
+# ----------------------------------------------------------------
+# --------------- LycheeAI live asset ----------------------------
+# ----------------------------------------------------------------
+
 
 @configclass
 class SoArm100ReachEnvCfg(ReachEnvCfg):
@@ -63,3 +67,48 @@ def __post_init__(self):
         self.scene.env_spacing = 2.5
         # disable randomization for play
         self.observations.policy.enable_corruption = False
+
+
+# ----------------------------------------------------------------
+# --------------- ROSCON ES 2025 asset ---------------------------
+# ----------------------------------------------------------------
+
+
+@configclass
+class SoArm100ReachRosConEnvCfg(ReachEnvCfg):
+    def __post_init__(self):
+        # post init of parent
+        super().__post_init__()
+
+        # switch robot to franka
+        self.scene.robot = SO_ARM100_ROSCON_CFG.replace(prim_path="{ENV_REGEX_NS}/Robot")
+        # override rewards
+        self.rewards.end_effector_position_tracking.params["asset_cfg"].body_names = ["wrist_2_link"]
+        self.rewards.end_effector_position_tracking_fine_grained.params["asset_cfg"].body_names = ["wrist_2_link"]
+        self.rewards.end_effector_orientation_tracking.params["asset_cfg"].body_names = ["wrist_2_link"]
+
+        self.rewards.end_effector_orientation_tracking.weight = 0.0
+
+        # override actions
+        self.actions.arm_action = mdp.JointPositionActionCfg(
+            asset_name="robot",
+            joint_names=[".*"],
+            scale=0.5,
+            use_default_offset=True,
+        )
+        # override command generator body
+        # end-effector is along z-direction
+        self.commands.ee_pose.body_name = ["wrist_2_link"]
+        # self.commands.ee_pose.ranges.pitch = (math.pi, math.pi)
+
+
+@configclass
+class SoArm100ReachRosConEnvCfg_PLAY(SoArm100ReachEnvCfg):
+    def __post_init__(self):
+        # post init of parent
+        super().__post_init__()
+        # make a smaller scene for play
+        self.scene.num_envs = 50
+        self.scene.env_spacing = 2.5
+        # disable randomization for play
+        self.observations.policy.enable_corruption = False