feat: introduce inference engine strategy

examples/phone_to_so100/record.py

@@ -16,14 +16,29 @@
 
 from lerobot.cameras.opencv import OpenCVCameraConfig
 from lerobot.common.control_utils import init_keyboard_listener
-from lerobot.datasets import LeRobotDataset
-from lerobot.processor import make_default_processors
+from lerobot.datasets import LeRobotDataset, aggregate_pipeline_dataset_features, create_initial_features
+from lerobot.model.kinematics import RobotKinematics
+from lerobot.processor import (
+    RobotProcessorPipeline,
+    observation_to_transition,
+    robot_action_observation_to_transition,
+    transition_to_observation,
+    transition_to_robot_action,
+)
 from lerobot.robots.so_follower import SO100Follower, SO100FollowerConfig
+from lerobot.robots.so_follower.robot_kinematic_processor import (
+    EEBoundsAndSafety,
+    EEReferenceAndDelta,
+    ForwardKinematicsJointsToEE,
+    GripperVelocityToJoint,
+    InverseKinematicsEEToJoints,
+)
 from lerobot.scripts.lerobot_record import record_loop
 from lerobot.teleoperators.phone import Phone, PhoneConfig
 from lerobot.teleoperators.phone.config_phone import PhoneOS
-from lerobot.utils.constants import ACTION, OBS_STR
-from lerobot.utils.feature_utils import hw_to_dataset_features
+from lerobot.teleoperators.phone.phone_processor import MapPhoneActionToRobotAction
+from lerobot.types import RobotAction, RobotObservation
+from lerobot.utils.feature_utils import combine_feature_dicts
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun
 
@@ -50,16 +65,77 @@ def main():
     robot = SO100Follower(robot_config)
     phone = Phone(teleop_config)
 
-    # Configure the dataset features
-    action_features = hw_to_dataset_features(robot.action_features, ACTION)
-    obs_features = hw_to_dataset_features(robot.observation_features, OBS_STR)
-    dataset_features = {**action_features, **obs_features}
+    # NOTE: It is highly recommended to use the urdf in the SO-ARM100 repo:
+    #   https://github.com/TheRobotStudio/SO-ARM100/blob/main/Simulation/SO101/so101_new_calib.urdf
+    kinematics_solver = RobotKinematics(
+        urdf_path="./SO101/so101_new_calib.urdf",
+        target_frame_name="gripper_frame_link",
+        joint_names=list(robot.bus.motors.keys()),
+    )
 
-    # Create the dataset
+    # Build pipeline to convert phone action to EE action (with gripper velocity mapped to joint).
+    phone_to_robot_ee_pose_processor = RobotProcessorPipeline[
+        tuple[RobotAction, RobotObservation], RobotAction
+    ](
+        steps=[
+            MapPhoneActionToRobotAction(platform=teleop_config.phone_os),
+            EEReferenceAndDelta(
+                kinematics=kinematics_solver,
+                end_effector_step_sizes={"x": 0.5, "y": 0.5, "z": 0.5},
+                motor_names=list(robot.bus.motors.keys()),
+                use_latched_reference=True,
+            ),
+            EEBoundsAndSafety(
+                end_effector_bounds={"min": [-1.0, -1.0, -1.0], "max": [1.0, 1.0, 1.0]},
+                max_ee_step_m=0.20,
+            ),
+            GripperVelocityToJoint(speed_factor=20.0),
+        ],
+        to_transition=robot_action_observation_to_transition,
+        to_output=transition_to_robot_action,
+    )
+
+    # Build pipeline to convert EE action to joints action (IK).
+    robot_ee_to_joints_processor = RobotProcessorPipeline[tuple[RobotAction, RobotObservation], RobotAction](
+        steps=[
+            InverseKinematicsEEToJoints(
+                kinematics=kinematics_solver,
+                motor_names=list(robot.bus.motors.keys()),
+                initial_guess_current_joints=True,
+            ),
+        ],
+        to_transition=robot_action_observation_to_transition,
+        to_output=transition_to_robot_action,
+    )
+
+    # Build pipeline to convert joint observation to EE observation (FK).
+    robot_joints_to_ee_pose = RobotProcessorPipeline[RobotObservation, RobotObservation](
+        steps=[
+            ForwardKinematicsJointsToEE(
+                kinematics=kinematics_solver, motor_names=list(robot.bus.motors.keys())
+            )
+        ],
+        to_transition=observation_to_transition,
+        to_output=transition_to_observation,
+    )
+
+    # Create the dataset, deriving features from the pipelines so the on-disk schema
+    # matches exactly what the pipelines produce at runtime.
     dataset = LeRobotDataset.create(
         repo_id=HF_REPO_ID,
         fps=FPS,
-        features=dataset_features,
+        features=combine_feature_dicts(
+            aggregate_pipeline_dataset_features(
+                pipeline=phone_to_robot_ee_pose_processor,
+                initial_features=create_initial_features(action=phone.action_features),
+                use_videos=True,
+            ),
+            aggregate_pipeline_dataset_features(
+                pipeline=robot_joints_to_ee_pose,
+                initial_features=create_initial_features(observation=robot.observation_features),
+                use_videos=True,
+            ),
+        ),
         robot_type=robot.name,
         use_videos=True,
         image_writer_threads=4,
@@ -77,10 +153,6 @@ def main():
         if not robot.is_connected or not phone.is_connected:
             raise ValueError("Robot or teleop is not connected!")
 
-        teleop_action_processor, robot_action_processor, robot_observation_processor = (
-            make_default_processors()
-        )
-
         print("Starting record loop. Move your phone to teleoperate the robot...")
         episode_idx = 0
         while episode_idx < NUM_EPISODES and not events["stop_recording"]:
@@ -91,9 +163,9 @@ def main():
                 robot=robot,
                 events=events,
                 fps=FPS,
-                teleop_action_processor=teleop_action_processor,
-                robot_action_processor=robot_action_processor,
-                robot_observation_processor=robot_observation_processor,
+                teleop_action_processor=phone_to_robot_ee_pose_processor,
+                robot_action_processor=robot_ee_to_joints_processor,
+                robot_observation_processor=robot_joints_to_ee_pose,
                 teleop=phone,
                 dataset=dataset,
                 control_time_s=EPISODE_TIME_SEC,
@@ -110,9 +182,9 @@ def main():
                     robot=robot,
                     events=events,
                     fps=FPS,
-                    teleop_action_processor=teleop_action_processor,
-                    robot_action_processor=robot_action_processor,
-                    robot_observation_processor=robot_observation_processor,
+                    teleop_action_processor=phone_to_robot_ee_pose_processor,
+                    robot_action_processor=robot_ee_to_joints_processor,
+                    robot_observation_processor=robot_joints_to_ee_pose,
                     teleop=phone,
                     control_time_s=RESET_TIME_SEC,
                     single_task=TASK_DESCRIPTION,