tmp

UMich-CURLY · Sep 29, 2018 · f0a4c69 · f0a4c69
commit f0a4c69
Show file tree

Hide file tree

Showing 91 changed files with 3,824 additions and 0 deletions.
diff --git a/.idea/misc.xml b/.idea/misc.xml
diff --git a/.idea/modules.xml b/.idea/modules.xml
diff --git a/.idea/other.xml b/.idea/other.xml
diff --git a/.idea/vcs.xml b/.idea/vcs.xml
diff --git a/.idea/vehicle-motion-forecasting.iml b/.idea/vehicle-motion-forecasting.iml
diff --git a/.idea/workspace.xml b/.idea/workspace.xml
diff --git a/.ipynb_checkpoints/demo-checkpoint.ipynb b/.ipynb_checkpoints/demo-checkpoint.ipynb
@@ -0,0 +1,163 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 26,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import matplotlib.pyplot as plt\n",
+    "import mdp.offroad_grid as offroad_grid\n",
+    "import numpy as np\n",
+    "from network.hybrid_dilated import HybridDilated\n",
+    "from torch.autograd import Variable\n",
+    "import torch\n",
+    "from os.path import join\n",
+    "import scipy.io as sio\n",
+    "from loader.util import leastsq_circle, calc_sign"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 27,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/home/yf/git_repo/vehicle-motion-forecasting/network/hybrid_dilated.py:59: UserWarning: nn.init.kaiming_normal is now deprecated in favor of nn.init.kaiming_normal_.\n",
+      "  nn.init.kaiming_normal(mod.weight, a=0)\n"
+     ]
+    }
+   ],
+   "source": [
+    "# initialize parameters\n",
+    "grid_size = 80\n",
+    "discount = 0.9\n",
+    "model = offroad_grid.OffroadGrid(grid_size, discount)\n",
+    "n_states = model.n_states\n",
+    "n_actions = model.n_actions\n",
+    "\n",
+    "net = HybridDilated(feat_out_size=25, regression_hidden_size=64)\n",
+    "net.init_weights()\n",
+    "net.load_state_dict(torch.load(join('example_data', 'example_weights.pth'))['net_state'])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 28,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def load(grid_size):\n",
+    "    \"\"\" load sample demo input data\"\"\"\n",
+    "    mean_std = sio.loadmat(join('example_data', 'data_mean_std.mat'))\n",
+    "    data_mat = sio.loadmat(join('example_data', 'demo_input.mat'))\n",
+    "    feat = data_mat['feat']\n",
+    "    # pre-process environment input\n",
+    "    feat[0] = (feat[0] - np.mean(feat[0])) / np.std(feat[0]) # normalize max-height feature locally (w.r.t. robot frame)\n",
+    "    feat[1] = (feat[1] - mean_std['variance_mean']) / mean_std['variance_std']\n",
+    "    feat[2] = (feat[2] - mean_std['red_mean']) / mean_std['red_std']\n",
+    "    feat[3] = (feat[3] - mean_std['green_mean']) / mean_std['green_std']\n",
+    "    feat[4] = (feat[4] - mean_std['blue_mean']) / mean_std['blue_std']\n",
+    "    \n",
+    "    # pre-process kinematic input\n",
+    "    past_traj, future_traj = data_mat['past_traj'], data_mat['future_traj']\n",
+    "    x, y = past_traj[:, 0], past_traj[:, 1]\n",
+    "    xc, yc, r, _ = leastsq_circle(x, y)\n",
+    "    curve_sign = calc_sign(x[0], y[0], x[-1], y[-1], xc, yc)\n",
+    "    kappa = 1.0 / r * curve_sign * 10.0 # 10.0 is empirically selected by observing the histogram\n",
+    "    feat = np.vstack((feat, np.full((1, grid_size, grid_size), kappa, dtype=np.float)))\n",
+    "    \n",
+    "    normalization = 0.5 * grid_size # 0.5*grid_size is used for normalize vx, vy, coordinate layers\n",
+    "    vx = (past_traj[-1, 0] - past_traj[0, 0]) / normalization\n",
+    "    vy = (past_traj[-1, 1] - past_traj[0 ,1]) / normalization\n",
+    "    feat = np.vstack((feat, np.full((1, grid_size, grid_size), vx, dtype=np.float)))\n",
+    "    feat = np.vstack((feat, np.full((1, grid_size, grid_size), vy, dtype=np.float)))\n",
+    "\n",
+    "    # coordinate layer\n",
+    "    center_idx = grid_size / 2\n",
+    "    delta_x_layer = np.zeros((1, grid_size, grid_size), dtype=np.float)\n",
+    "    delta_y_layer = delta_x_layer.copy()\n",
+    "    for x in range(grid_size):\n",
+    "        for y in range(grid_size):\n",
+    "            delta_x_layer[0, x, y] = x - center_idx\n",
+    "            delta_y_layer[0, x, y] = y - center_idx\n",
+    "    feat = np.vstack((feat, delta_x_layer / normalization))\n",
+    "    feat = np.vstack((feat, delta_y_layer / normalization))\n",
+    "    \n",
+    "    return feat, past_traj, future_traj"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 31,
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "RuntimeError",
+     "evalue": "Expected object of type torch.DoubleTensor but found type torch.FloatTensor for argument #2 'weight'",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
+      "\u001b[0;32m<ipython-input-31-bc765b89d6b9>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0mfeat\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexpand_dims\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0mfeat_var\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mVariable\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfrom_numpy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtype\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mDoubleTensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 4\u001b[0;31m \u001b[0mr_var\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnet\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat_var\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
+      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m    489\u001b[0m             \u001b[0mresult\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_slow_forward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    490\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 491\u001b[0;31m             \u001b[0mresult\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    492\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mhook\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_forward_hooks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mvalues\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    493\u001b[0m             \u001b[0mhook_result\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mhook\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mresult\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
+      "\u001b[0;32m~/git_repo/vehicle-motion-forecasting/network/hybrid_dilated.py\u001b[0m in \u001b[0;36mforward\u001b[0;34m(self, x)\u001b[0m\n\u001b[1;32m     44\u001b[0m         \u001b[0;31m# geometric and semantic feature extraction\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     45\u001b[0m         \u001b[0mfeat_in\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfeat_in_size\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 46\u001b[0;31m         \u001b[0mfeat_out\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfeat_block\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat_in\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     47\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     48\u001b[0m         \u001b[0mkinematic_in\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfeat_out\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m5\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdim\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
+      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m    489\u001b[0m             \u001b[0mresult\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_slow_forward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    490\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 491\u001b[0;31m             \u001b[0mresult\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    492\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mhook\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_forward_hooks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mvalues\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    493\u001b[0m             \u001b[0mhook_result\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mhook\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mresult\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
+      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torch/nn/modules/container.py\u001b[0m in \u001b[0;36mforward\u001b[0;34m(self, input)\u001b[0m\n\u001b[1;32m     89\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minput\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     90\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mmodule\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_modules\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mvalues\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 91\u001b[0;31m             \u001b[0minput\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodule\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     92\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0minput\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     93\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
+      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m    489\u001b[0m             \u001b[0mresult\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_slow_forward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    490\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 491\u001b[0;31m             \u001b[0mresult\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    492\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mhook\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_forward_hooks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mvalues\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    493\u001b[0m             \u001b[0mhook_result\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mhook\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mresult\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
+      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torch/nn/modules/conv.py\u001b[0m in \u001b[0;36mforward\u001b[0;34m(self, input)\u001b[0m\n\u001b[1;32m    299\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minput\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    300\u001b[0m         return F.conv2d(input, self.weight, self.bias, self.stride,\n\u001b[0;32m--> 301\u001b[0;31m                         self.padding, self.dilation, self.groups)\n\u001b[0m\u001b[1;32m    302\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    303\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
+      "\u001b[0;31mRuntimeError\u001b[0m: Expected object of type torch.DoubleTensor but found type torch.FloatTensor for argument #2 'weight'"
+     ]
+    }
+   ],
+   "source": [
+    "feat, past_traj, future_traj = load(grid_size)\n",
+    "feat = np.expand_dims(feat, axis=0)\n",
+    "feat_var = Variable(torch.from_numpy(feat).type(torch.DoubleTensor))\n",
+    "r_var = net(feat_var)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 30,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "torch.from_numpy??"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.6.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
diff --git a/README.md b/README.md
@@ -0,0 +1,38 @@
+## branch说明
+### v0
+- 用FCN网络,有spatial learning
+- 输入过去的trajectory和全部点云地图上的特征(height,var,red,green,blue)
+
+### v1
+- 根据过去的trajectory提取方向和速度信息.方向信息又sin和cos来表示.
+- 把速度,sin,cos信息分别放成一个layer提供给network.(速度layer的每个grid都是同一个速度值)
+
+### v2
+- 根据过去的trajectory(方向,速度),预测一个简单的轨迹,直接当做一层feature提供给网络
+
+### v3
+- 按照activity forecasting里的做法,每次value iteration的时候,把goal state的value重置为0
+
+### v4
+- 把速度,方向,和位置相关的信息,直接输入到最后几层做regression的layer
+- 前面几层卷积层用来提取和geometry,RGB相关的feature
+
+### v5
+- 重新整理训练数据,平衡拐弯和直线的比例
+
+### v6
+- 添加新的训练数据 (平衡在岔路口的选择,在宽阔路段的S形drive)
+  - 6.1, 加pre-train
+    - 6.11, 正常的Q; discount 0.95
+    - 6.12, kinematic信息, x,y从0到1, 没负数
+    - 6.13, kinematic信息, x,y从-2到2; Q乘10; discount 0.95
+    - 6.14, kinematic信息, x,y从-1到1; Q乘20; discount 0.95
+  - 6.2, 增加narrow trail的数据
+    - 6.23, (修正)kinematic信息, x,y从-1到1, feat_out=25, regression_hidden_size=64
+    - 6.25, 删去training里面的S型
+    - 6.26, 重新train
+    - 6.27, 用tangent
+  - 6.3, random加uniform环境,只有kinematics, random=0.5, tanget
+    - 6.31, random=0.2, tangent
+    - 6.32, random=0.2, no tagent
+    - 6.33, random=0.3, 删S型(修正)