Add FeLT data & fix ml examples

examples/data/felt/Readme.txt

@@ -0,0 +1,24 @@
+Data (csv format) is organized in folders first by task then by trial. Three trial folders are included for each of the following tasks:
+- raster: wiping the surface of a bamboo cutting board side to side as the hand  travels from the bottom ot the top or top to bottom
+- circle: wiping the surface of a bamboo cutting board in small counterclockwise circles as the hand covers travels around different parts in a general counterclockwise trajectory.
+- spiral: wiping the surface of a bamboo cutting board starting in its center and spiraling ountward, then spiraling inwards once coming closer to the edge (without reaching it)
+- edgeC: following the edge of a round plastic bowl, with half the sensor making contact
+- edgeR: following the edge of a rectangular bamboo cutting board, with half the sensor making contact
+Each trial is a recording of approximately 55 continuous seconds.
+
+ Each trial folder contains data files corresponding to the following tasks:
+- wipe_{task}_mocap_hand: motion capture (mocap) data for the hand that’s holding the sensor (3D coordinates of 8 markers = 16 columns). Columns are grouped by threes (i.e. x-y-z, x-y-z,...)
+- wipe_{task}_mocap_object: mocap data for the object (16 columns as above)
+- wipe_{task}_sensor: comprehensive tactile sensor data (4 deflection directions of 9 taxels = 36 columns)
+- wipe_{task}_sensor_comb: tactile sensor data grouped along different directions  (4 cardinal directions, 2 rotational (clockwise and counterclockwise), and overall pressure = 7 columns)
+- wipe_{task}_sensor_partial_comb: portions of the taxels are combined, to differentiate between differentiate partial contacts/interactions
+
+Each data file includes column labels (first row) and data (rest of the rows). Each row is a frame (sample) of data, sampled at 60Hz.
+
+Columns headings for the sensor data are defined as follows:
+- wipe_{task}_sensor: taxel_#_left,taxel_#_up,taxel_#_right,taxel_#_down, where # ranges from 1 to 9. Taxels are numbered with 1 being the bottom left, 2 is bottom center, 3 is bottom right, moving up to 9 being top right.
+- wipe_{task}_sensor_comb: [taxels_up,taxels_down,taxels_right,taxels_left,taxels_clockwise,taxels_counterclockwise,taxels_pressure]. See Fig. 5 in https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8990001 for a visual explanation of how the taxels are summed for the linear and rotational columns.
+- wipe_{task}_sensor_partial_comb: [taxel_left_up] corresponds to measuring the upward deflection of the left three taxels. Another example, [taxel_down_right] corresponds to measuring rightward deflection of the bottom three taxels. [taxels_{side}_pressure] corresponds to the overall pressure (sum of all deflection directions) of the corresponding sensor portion (defined by side: up/down/left/right).
+
+Columni headings for the mocap data is in the following order:
+- [marker_#_x,marker_#_y,marker_#_z,marker_#_x,...] where # ranges from 1 to 8 (markers).

examples/learning_ml/example_real_human_tpgmm_birp.py → examples/learning_ml/example_birp.py

@@ -1,8 +1,9 @@
 """
-TP-GMM learning from real human demonstration data
+BiRP
 =================================================
 
-This example shows how to learn a TP-GMM from real human demonstration data.
+The code for the paper "BiRP: BiRP: Learning Robot Generalized Bimanual Coordination using Relative Parameterization
+Method on Human Demonstration" by Junjia LIU, et al. (https://arxiv.org/abs/2307.05933)
 """
 
 from isaacgym import gymutil

examples/learning_ml/example_felt.py

@@ -0,0 +1,37 @@
+"""
+FeLT
+=================
+
+The coder for the paper "FeLT: Fully Tactile-driven Robot Plate Cleaning Skill Learning from Human Demonstration
+ with Tactile Sensor" by Junjia LIU, et al.
+"""
+import numpy as np
+
+import rofunc as rf
+
+
+# --- Data processing ---
+def data_process(dat_path):
+    raw_demo = np.load('../../data/LFD_ML/LeftHand.npy')
+
+
+demos_x = data_process('../data/LFD_ML/LeftHand.npy')
+
+# --- TP-GMM ---
+# Define the task parameters
+start_xdx = [demos_x[i][0] for i in range(len(demos_x))]  # TODO: change to xdx
+end_xdx = [demos_x[i][-1] for i in range(len(demos_x))]
+task_params = {'frame_origins': [start_xdx, end_xdx], 'frame_names': ['start', 'end']}
+# Fit the model
+Repr = rf.ml.TPGMM(demos_x, task_params, plot=True)
+model = Repr.fit()
+
+# Reproductions for the same situations
+traj, _ = Repr.reproduce(model, show_demo_idx=2)
+
+# Reproductions for new situations: set the endpoint as the start point to make a cycled motion
+ref_demo_idx = 2
+start_xdx = [Repr.demos_xdx[ref_demo_idx][0]]
+end_xdx = [Repr.demos_xdx[ref_demo_idx][0]]
+Repr.task_params = {'frame_origins': [start_xdx, end_xdx], 'frame_names': ['start', 'end']}
+traj, _ = Repr.generate(model, ref_demo_idx)

examples/learning_ml/example_tpgmm.py

@@ -8,7 +8,7 @@
 
 import rofunc as rf
 
-raw_demo = np.load('../../data/LFD_ML/LeftHand.npy')
+raw_demo = np.load('../data/LFD_ML/LeftHand.npy')
 demos_x = [raw_demo[500:635, :], raw_demo[635:770, :], raw_demo[770:905, :]]
 
 # --- TP-GMM ---

examples/learning_ml/example_tpgmmbi.py

@@ -7,8 +7,8 @@
 import numpy as np
 import rofunc as rf
 
-left_raw_demo = np.load('../../data/LFD_ML/LeftHand.npy')
-right_raw_demo = np.load('../../data/LFD_ML/RightHand.npy')
+left_raw_demo = np.load('../data/LFD_ML/LeftHand.npy')
+right_raw_demo = np.load('../data/LFD_ML/RightHand.npy')
 demos_left_x = [left_raw_demo[500:635, :], left_raw_demo[635:770, :], left_raw_demo[770:905, :]]
 demos_right_x = [right_raw_demo[500:635, :], right_raw_demo[635:770, :], right_raw_demo[770:905, :]]
 

examples/learning_ml/example_tpgmr.py

@@ -8,7 +8,7 @@
 
 import rofunc as rf
 
-raw_demo = np.load('../../data/LFD_ML/LeftHand.npy')
+raw_demo = np.load('../data/LFD_ML/LeftHand.npy')
 demos_x = [raw_demo[500:635, :], raw_demo[635:770, :], raw_demo[770:905, :]]
 
 # --- TP-GMR ---