Fetch mobile (#188)

* Progress: Fetch mobile base

* fix

* fix

* unfinished: integrate into gpu sim stuff

* Fix: gpu sim compatibility issues

* Fix: fetch sensor cfg

* Fix: pd_ee_pose controller ik, env step cpu sim output, utils, fetch pose

* Fix: fetch is_grasping cpu sim

mani_skill2/agents/controllers/base_controller.py

@@ -206,12 +206,7 @@ def _initialize_joints(self):
 
     def _assert_fully_actuated(self):
         active_joints = self.articulation.get_active_joints()
-        if len(active_joints) != len(self.joints) or not np.all(
-            [
-                active_joint == joint
-                for active_joint, joint in zip(active_joints, self.joints)
-            ]
-        ):
+        if len(active_joints) != len(self.joints) or set(active_joints) != set(self.joints):
             print("active_joints:", [x.name for x in active_joints])
             print("controlled_joints:", [x.name for x in self.joints])
             raise AssertionError("{} is not fully actuated".format(self.articulation))

mani_skill2/agents/controllers/pd_base_vel.py

@@ -1,4 +1,5 @@
 import numpy as np
+import torch
 
 from mani_skill2.utils.geometry import rotate_2d_vec_by_angle
 
@@ -18,14 +19,21 @@ def _initialize_action_space(self):
     def set_action(self, action: np.ndarray):
         action = self._preprocess_action(action)
 
+        # TODO (arth): add support for batched qpos and gpu sim
+        if isinstance(self.qpos, torch.Tensor):
+            qpos = self.qpos.detach().cpu().numpy()
+        qpos = qpos[0]
+        if isinstance(action, torch.Tensor):
+            action = action.detach().cpu().numpy()
+
         # Convert to ego-centric action
         # Assume the 3rd DoF stands for orientation
-        ori = self.qpos[2]
+        ori = qpos[2]
         vel = rotate_2d_vec_by_angle(action[:2], ori)
         new_action = np.hstack([vel, action[2:]])
 
         for i, joint in enumerate(self.joints):
-            joint.set_drive_velocity_target(new_action[i])
+            joint.set_drive_velocity_target(np.array([new_action[i]]))
 
 
 class PDBaseVelControllerConfig(PDJointVelControllerConfig):

mani_skill2/agents/controllers/pd_ee_pose.py

@@ -8,7 +8,7 @@
 from scipy.spatial.transform import Rotation
 
 from mani_skill2.utils.common import clip_and_scale_action
-from mani_skill2.utils.sapien_utils import get_obj_by_name
+from mani_skill2.utils.sapien_utils import get_obj_by_name, to_numpy, to_tensor
 from mani_skill2.utils.structs.pose import vectorize_pose
 
 from .base_controller import BaseController, ControllerConfig
@@ -64,15 +64,17 @@ def reset(self):
 
     def compute_ik(self, target_pose, max_iterations=100):
         # Assume the target pose is defined in the base frame
+        # TODO (arth): currently ik only supports cpu, so input/output is managed as such
+        #       in future, need to change input/output processing per gpu implementation
         result, success, error = self.pmodel.compute_inverse_kinematics(
             self.ee_link_idx,
-            target_pose,
-            initial_qpos=self.articulation.get_qpos(),
+            target_pose.sp,
+            initial_qpos=to_numpy(self.articulation.get_qpos()).squeeze(0),
             active_qmask=self.qmask,
             max_iterations=max_iterations,
         )
         if success:
-            return result[self.joint_indices]
+            return to_tensor([result[self.joint_indices]])
         else:
             return None
 

mani_skill2/agents/robots/fetch/__init__.py

@@ -1 +1 @@
-from .fetch import Fetch
+from .fetch import Fetch, FETCH_UNIQUE_COLLISION_BIT

mani_skill2/agents/robots/fetch/fetch.py

@@ -1,21 +1,31 @@
 from copy import deepcopy
+from typing import Dict, Tuple
 
 import numpy as np
 import sapien
 import sapien.physx as physx
+import torch
 
 from mani_skill2 import PACKAGE_ASSET_DIR
 from mani_skill2.agents.base_agent import BaseAgent
 from mani_skill2.agents.controllers import *
 from mani_skill2.sensors.camera import CameraConfig
-from mani_skill2.utils.common import np_compute_angle_between
+from mani_skill2.utils.common import compute_angle_between, np_compute_angle_between
 from mani_skill2.utils.sapien_utils import (
     compute_total_impulse,
     get_actor_contacts,
     get_obj_by_name,
     get_pairwise_contact_impulse,
 )
+from mani_skill2.utils.structs.actor import Actor
+from mani_skill2.utils.sapien_utils import to_tensor
+from mani_skill2.utils.structs.base import BaseStruct
+from mani_skill2.utils.structs.joint import Joint
+from mani_skill2.utils.structs.link import Link
+from mani_skill2.utils.structs.pose import Pose
+from mani_skill2.utils.structs.types import Array
 
+FETCH_UNIQUE_COLLISION_BIT = 1 << 30
 
 class Fetch(BaseAgent):
     uid = "fetch"
@@ -33,6 +43,19 @@ class Fetch(BaseAgent):
             ),
         ),
     )
+    sensor_configs = [
+        CameraConfig(
+            uid="fetch_head",
+            p=[0, 0, 0],
+            q=[1, 0, 0, 0],
+            width=128,
+            height=128,
+            fov=1.57,
+            near=0.01,
+            far=10,
+            entity_uid="head_camera_link",
+        )
+    ]
 
     def __init__(self, *args, **kwargs):
         self.arm_joint_names = [
@@ -67,6 +90,12 @@ def __init__(self, *args, **kwargs):
         self.body_damping = 1e2
         self.body_force_limit = 100
 
+        self.base_joint_names = [
+            "root_x_axis_joint",
+            "root_y_axis_joint",
+            "root_z_rotation_joint",
+        ]
+
         super().__init__(*args, **kwargs)
 
     @property
@@ -181,110 +210,164 @@ def controller_configs(self):
             normalize_action=False,
         )
 
+        # -------------------------------------------------------------------------- #
+        # Base
+        # -------------------------------------------------------------------------- #
+        base_pd_joint_vel = PDBaseVelControllerConfig(
+            self.base_joint_names,
+            lower=[-0.5, -0.5, -3.14],
+            upper=[0.5, 0.5, 3.14],
+            damping=1000,
+            force_limit=500,
+        )
+
+
         controller_configs = dict(
             pd_joint_delta_pos=dict(
-                arm=arm_pd_joint_delta_pos,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_joint_delta_pos, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_joint_pos=dict(
-                arm=arm_pd_joint_pos,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_joint_pos, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_ee_delta_pos=dict(
-                arm=arm_pd_ee_delta_pos,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_ee_delta_pos, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_ee_delta_pose=dict(
-                arm=arm_pd_ee_delta_pose,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_ee_delta_pose, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_ee_delta_pose_align=dict(
-                arm=arm_pd_ee_delta_pose_align,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_ee_delta_pose_align, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             # TODO(jigu): how to add boundaries for the following controllers
             pd_joint_target_delta_pos=dict(
-                arm=arm_pd_joint_target_delta_pos,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_joint_target_delta_pos, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_ee_target_delta_pos=dict(
-                arm=arm_pd_ee_target_delta_pos,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_ee_target_delta_pos, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_ee_target_delta_pose=dict(
-                arm=arm_pd_ee_target_delta_pose,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_ee_target_delta_pose, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             # Caution to use the following controllers
             pd_joint_vel=dict(
-                arm=arm_pd_joint_vel,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_joint_vel, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_joint_pos_vel=dict(
-                arm=arm_pd_joint_pos_vel,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_joint_pos_vel, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
             pd_joint_delta_pos_vel=dict(
-                arm=arm_pd_joint_delta_pos_vel,
-                gripper=gripper_pd_joint_pos,
-                body=body_pd_joint_pos,
+                arm=arm_pd_joint_delta_pos_vel, gripper=gripper_pd_joint_pos, body=body_pd_joint_pos, base=base_pd_joint_vel,
             ),
         )
 
         # Make a deepcopy in case users modify any config
         return deepcopy_dict(controller_configs)
 
     def _after_init(self):
-        self.finger1_link = get_obj_by_name(
+        self.finger1_link: Link = get_obj_by_name(
             self.robot.get_links(), "l_gripper_finger_link"
         )
-        self.finger2_link = get_obj_by_name(
+        self.finger2_link: Link = get_obj_by_name(
             self.robot.get_links(), "r_gripper_finger_link"
         )
-        self.tcp = get_obj_by_name(self.robot.get_links(), self.ee_link_name)
-
-    def is_grasping(self, object: sapien.Entity = None, min_impulse=1e-6, max_angle=85):
-        contacts = self.scene.get_contacts()
-        if object is None:
-            finger1_contacts = get_actor_contacts(contacts, self.finger1_link)
-            finger2_contacts = get_actor_contacts(contacts, self.finger2_link)
-            return (
-                np.linalg.norm(compute_total_impulse(finger1_contacts)) >= min_impulse
-                and np.linalg.norm(compute_total_impulse(finger2_contacts))
-                >= min_impulse
-            )
-        else:
-            limpulse = get_pairwise_contact_impulse(contacts, self.finger1_link, object)
-            rimpulse = get_pairwise_contact_impulse(contacts, self.finger2_link, object)
+        self.tcp: Link = get_obj_by_name(
+            self.robot.get_links(), self.ee_link_name
+        )
 
-            # direction to open the gripper
-            ldirection = -self.finger1_link.pose.to_transformation_matrix()[:3, 1]
-            rdirection = self.finger2_link.pose.to_transformation_matrix()[:3, 1]
+        self.base_link: Link = get_obj_by_name(
+            self.robot.get_links(), "base_link"
+        )
+        self.l_wheel_link: Link = get_obj_by_name(
+            self.robot.get_links(), "l_wheel_link"
+        )
+        self.r_wheel_link: Link = get_obj_by_name(
+            self.robot.get_links(), "r_wheel_link"
+        )
+        for link in [self.base_link, self.l_wheel_link, self.r_wheel_link]:
+            cs = link._bodies[0].get_collision_shapes()[0]
+            cg = cs.get_collision_groups()
+            cg[2] = FETCH_UNIQUE_COLLISION_BIT
+            cs.set_collision_groups(cg)
 
-            # angle between impulse and open direction
-            langle = np_compute_angle_between(ldirection, limpulse)
-            rangle = np_compute_angle_between(rdirection, rimpulse)
+        self.queries: Dict[str, Tuple[physx.PhysxGpuContactQuery, Tuple[int]]] = dict()
 
-            lflag = (
-                np.linalg.norm(limpulse) >= min_impulse
-                and np.rad2deg(langle) <= max_angle
+    def is_grasping(self, object: Actor = None, min_impulse=1e-6, max_angle=85):
+        # TODO (stao): is_grasping code needs to be updated for new GPU sim
+        if physx.is_gpu_enabled():
+            if object.name not in self.queries:
+                body_pairs = list(zip(self.finger1_link._bodies, object._bodies))
+                body_pairs += list(zip(self.finger2_link._bodies, object._bodies))
+                self.queries[object.name] = (
+                    self.scene.px.gpu_create_contact_query(body_pairs),
+                    (len(object._bodies), 3),
+                )
+                print(f"Create query for Fetch grasp({object.name})")
+            query, contacts_shape = self.queries[object.name]
+            self.scene.px.gpu_query_contacts(query)
+            # query.cuda_contacts # (num_unique_pairs * num_envs, 3)
+            contacts = query.cuda_contacts.clone().reshape((-1, *contacts_shape))
+            lforce = torch.linalg.norm(contacts[0], axis=1)
+            rforce = torch.linalg.norm(contacts[1], axis=1)
+
+            # NOTE (stao): 0.5 * time_step is a decent value when tested on a pick cube task.
+            min_force = 0.5 * self.scene.px.timestep
+
+            # direction to open the gripper
+            ldirection = -self.finger1_link.pose.to_transformation_matrix()[..., :3, 1]
+            rdirection = self.finger2_link.pose.to_transformation_matrix()[..., :3, 1]
+            langle = compute_angle_between(ldirection, contacts[0])
+            rangle = compute_angle_between(rdirection, contacts[1])
+            lflag = torch.logical_and(
+                lforce >= min_force, torch.rad2deg(langle) <= max_angle
             )
-            rflag = (
-                np.linalg.norm(rimpulse) >= min_impulse
-                and np.rad2deg(rangle) <= max_angle
+            rflag = torch.logical_and(
+                rforce >= min_force, torch.rad2deg(rangle) <= max_angle
             )
 
-            return all([lflag, rflag])
+            return torch.logical_and(lflag, rflag)
+        else:
+            contacts = self.scene.get_contacts()
+
+            if object is None:
+                finger1_contacts = get_actor_contacts(contacts, self.finger1_link._bodies[0].entity)
+                finger2_contacts = get_actor_contacts(contacts, self.finger2_link._bodies[0].entity)
+                return (
+                    np.linalg.norm(compute_total_impulse(finger1_contacts))
+                    >= min_impulse
+                    and np.linalg.norm(compute_total_impulse(finger2_contacts))
+                    >= min_impulse
+                )
+            else:
+                limpulse = get_pairwise_contact_impulse(
+                    contacts, self.finger1_link._bodies[0].entity, object._bodies[0].entity
+                )
+                rimpulse = get_pairwise_contact_impulse(
+                    contacts, self.finger2_link._bodies[0].entity, object._bodies[0].entity
+                )
+
+                # direction to open the gripper
+                ldirection = -self.finger1_link.pose.to_transformation_matrix()[
+                    ..., :3, 1
+                ]
+                rdirection = self.finger2_link.pose.to_transformation_matrix()[
+                    ..., :3, 1
+                ]
+
+                # TODO Convert this to batched code
+                # angle between impulse and open direction
+                langle = np_compute_angle_between(ldirection[0], limpulse)
+                rangle = np_compute_angle_between(rdirection[0], rimpulse)
+
+                lflag = (
+                    np.linalg.norm(limpulse) >= min_impulse
+                    and np.rad2deg(langle) <= max_angle
+                )
+                rflag = (
+                    np.linalg.norm(rimpulse) >= min_impulse
+                    and np.rad2deg(rangle) <= max_angle
+                )
+
+                return all([lflag, rflag])
 
     @staticmethod
     def build_grasp_pose(approaching, closing, center):
@@ -297,23 +380,9 @@ def build_grasp_pose(approaching, closing, center):
         T[:3, :3] = np.stack([ortho, closing, approaching], axis=1)
         T[:3, 3] = center
         return sapien.Pose(T)
-
-    @property
-    def sensor_configs(self):
-        return [
-            CameraConfig(
-                uid="fetch_head",
-                p=[0, 0, 0],
-                q=[0.9238795, 0, 0.3826834, 0],
-                width=128,
-                height=128,
-                fov=1.57,
-                near=0.01,
-                far=10,
-                entity_uid="head_camera_link",
-            )
-        ]
-
+
     @property
-    def tcp_pose_p(self):
-        return (self.finger1_link.pose.p + self.finger2_link.pose.p) / 2
+    def tcp_pose(self) -> Pose:
+        p = (self.finger1_link.pose.p + self.finger2_link.pose.p) / 2
+        q = (self.finger1_link.pose.q + self.finger2_link.pose.q) / 2
+        return Pose.create_from_pq(p=p, q=q)