Add video2robot + MOVIN Studio to pipeline, fix Motion-Player-ROS links, add V_Rocamena, 60 clips

README.md

@@ -10,7 +10,7 @@ license: cc-by-4.0
 
 # G1 Moves
 
-59 motion capture clips for the Unitree G1 humanoid robot (29 DOF, mode 15). Dance, karate, fencing, and more — captured with MOVIN TRACIN, retargeted with Motion-Player-ROS, trained with mjlab, and deployed via RoboJuDo.
+60 motion capture clips for the Unitree G1 humanoid robot (29 DOF, mode 15). Dance, karate, fencing, and more — captured with MOVIN TRACIN, retargeted with Motion-Player-ROS, trained with mjlab, and deployed via RoboJuDo.
 
 Browse the interactive gallery above to view trained RL policies running in real-time in your browser via MuJoCo WASM + ONNX Runtime.
 
@@ -22,9 +22,11 @@ Browse the interactive gallery above to view trained RL policies running in real
 
 Motion is captured with the [MOVIN TRACIN](https://movin3d.com/product/movin-tracin/) markerless mocap system (LiDAR + vision, 60 FPS) and recorded in [MOVIN Studio](https://www.movin3d.com/studio). Output is BVH format with a 51-joint humanoid skeleton.
 
+Alternatively, motions can be extracted from any video (YouTube, phone, AI-generated) using [video2robot](https://github.com/AIM-Intelligence/video2robot), which runs PromptHMR pose extraction followed by [GMR](https://github.com/YanjieZe/GMR) inverse kinematics retargeting — no mocap hardware needed.
+
 ### 2. Retarget
 
-Each BVH clip is retargeted to the Unitree G1's 29-DOF joint space using per-frame inverse kinematics via [movin_sdk_python](https://movin3d.com/) and a custom [ROS 2 retargeting node](https://github.com/ExperientialTechnologies/Motion-Player-ROS). The node provides real-time preview with dual BVH skeleton + robot visualization in RViz, and supports both playback of pre-recorded `.pkl` files and live retargeting from MOVIN TRACIN data via OSC/UDP.
+Each BVH clip is retargeted to the Unitree G1's 29-DOF joint space using per-frame inverse kinematics via [MOVIN-SDK-Python](https://github.com/MOVIN3D/MOVIN-SDK-Python) and a custom [ROS 2 retargeting node](https://github.com/MOVIN3D/Motion-Player-ROS). The node provides real-time preview with dual BVH skeleton + robot visualization in RViz, and supports both playback of pre-recorded `.pkl` files and live retargeting from MOVIN TRACIN data via OSC/UDP.
 
 ```bash
 # Playback mode

data.json

@@ -1083,9 +1083,13 @@
         "training": {
           "gif": "karate/M_Move17/training/M_Move17_training.gif",
           "mp4": "karate/M_Move17/training/M_Move17_training.mp4"
+        },
+        "policy": {
+          "gif": "karate/M_Move17/policy/M_Move17_policy.gif",
+          "mp4": "karate/M_Move17/policy/M_Move17_policy.mp4"
         }
       },
-      "has_policy": false,
+      "has_policy": true,
       "has_onnx": true
     },
     {
@@ -1158,9 +1162,13 @@
         "training": {
           "gif": "karate/M_Move2/training/M_Move2_training.gif",
           "mp4": "karate/M_Move2/training/M_Move2_training.mp4"
+        },
+        "policy": {
+          "gif": "karate/M_Move2/policy/M_Move2_policy.gif",
+          "mp4": "karate/M_Move2/policy/M_Move2_policy.mp4"
         }
       },
-      "has_policy": false,
+      "has_policy": true,
       "has_onnx": true
     },
     {
@@ -1208,9 +1216,13 @@
         "training": {
           "gif": "karate/M_Move3/training/M_Move3_training.gif",
           "mp4": "karate/M_Move3/training/M_Move3_training.mp4"
+        },
+        "policy": {
+          "gif": "karate/M_Move3/policy/M_Move3_policy.gif",
+          "mp4": "karate/M_Move3/policy/M_Move3_policy.mp4"
         }
       },
-      "has_policy": false,
+      "has_policy": true,
       "has_onnx": true
     },
     {
@@ -1233,9 +1245,13 @@
         "training": {
           "gif": "karate/M_Move4/training/M_Move4_training.gif",
           "mp4": "karate/M_Move4/training/M_Move4_training.mp4"
+        },
+        "policy": {
+          "gif": "karate/M_Move4/policy/M_Move4_policy.gif",
+          "mp4": "karate/M_Move4/policy/M_Move4_policy.mp4"
         }
       },
-      "has_policy": false,
+      "has_policy": true,
       "has_onnx": true
     },
     {
@@ -1258,9 +1274,13 @@
         "training": {
           "gif": "karate/M_Move5/training/M_Move5_training.gif",
           "mp4": "karate/M_Move5/training/M_Move5_training.mp4"
+        },
+        "policy": {
+          "gif": "karate/M_Move5/policy/M_Move5_policy.gif",
+          "mp4": "karate/M_Move5/policy/M_Move5_policy.mp4"
         }
       },
-      "has_policy": false,
+      "has_policy": true,
       "has_onnx": true
     },
     {
@@ -1283,9 +1303,13 @@
         "training": {
           "gif": "karate/M_Move6/training/M_Move6_training.gif",
           "mp4": "karate/M_Move6/training/M_Move6_training.mp4"
+        },
+        "policy": {
+          "gif": "karate/M_Move6/policy/M_Move6_policy.gif",
+          "mp4": "karate/M_Move6/policy/M_Move6_policy.mp4"
         }
       },
-      "has_policy": false,
+      "has_policy": true,
       "has_onnx": true
     },
     {
@@ -1358,9 +1382,13 @@
         "training": {
           "gif": "karate/M_Move9/training/M_Move9_training.gif",
           "mp4": "karate/M_Move9/training/M_Move9_training.mp4"
+        },
+        "policy": {
+          "gif": "karate/M_Move9/policy/M_Move9_policy.gif",
+          "mp4": "karate/M_Move9/policy/M_Move9_policy.mp4"
         }
       },
-      "has_policy": false,
+      "has_policy": true,
       "has_onnx": true
     },
     {
@@ -1603,13 +1631,38 @@
       },
       "has_policy": true,
       "has_onnx": true
+    },
+    {
+      "id": "V_Rocamena",
+      "name": "Rocamena",
+      "category": "bonus",
+      "performer": "YouTube",
+      "duration": 9.5,
+      "fps": 30,
+      "frames": 285,
+      "stages": {
+        "capture": {
+          "gif": "bonus/V_Rocamena/capture/V_Rocamena.gif",
+          "mp4": "bonus/V_Rocamena/capture/V_Rocamena.mp4"
+        },
+        "retarget": {
+          "gif": "bonus/V_Rocamena/retarget/V_Rocamena_retarget.gif",
+          "mp4": "bonus/V_Rocamena/retarget/V_Rocamena_retarget.mp4"
+        },
+        "training": {
+          "gif": "bonus/V_Rocamena/training/V_Rocamena_training.gif",
+          "mp4": "bonus/V_Rocamena/training/V_Rocamena_training.mp4"
+        }
+      },
+      "has_policy": false,
+      "has_onnx": false
     }
   ],
   "stats": {
     "dance": 28,
     "karate": 27,
-    "bonus": 4,
-    "policies": 32,
-    "total": 59
+    "bonus": 5,
+    "policies": 39,
+    "total": 60
   }
 }

index.html

@@ -4,7 +4,7 @@
   <meta charset="UTF-8">
   <meta name="viewport" content="width=device-width, initial-scale=1.0">
   <title>G1 Moves — Motion Capture for Unitree G1</title>
-  <meta name="description" content="59 motion capture clips for the Unitree G1 humanoid robot. Dance, karate, and more — captured with MOVIN3D, trained with MuJoCo-Warp.">
+  <meta name="description" content="60 motion capture clips for the Unitree G1 humanoid robot. Dance, karate, and more — captured with MOVIN3D, trained with MuJoCo-Warp.">
 
   <link rel="preconnect" href="https://fonts.googleapis.com">
   <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
@@ -20,7 +20,7 @@
     <div class="hero-content">
       <img src="logo.png" alt="Experiential Technologies" class="logo" width="260">
       <h1 class="hero-title">G1 MOVES</h1>
-      <p class="hero-subtitle">59 motion capture clips for the Unitree G1 humanoid robot</p>
+      <p class="hero-subtitle">60 motion capture clips for the Unitree G1 humanoid robot</p>
 
       <div class="pipeline-strip">
         <div class="pipeline-stage">
@@ -93,13 +93,13 @@
       <h2>ABOUT</h2>
       <div class="about-columns">
         <div class="about-text">
-          <p>G1 Moves is a dataset of 59 motion capture clips captured with the
+          <p>G1 Moves is a dataset of 60 motion capture clips captured with the
           <a href="https://movin3d.com/" target="_blank" rel="noopener">MOVIN TRACIN</a>
           markerless capture system (LiDAR + vision, 60 FPS). Each clip is retargeted to the Unitree G1 humanoid robot
           (29 DOF, mode 15) using inverse kinematics via the
-          <a href="https://movin3d.com/" target="_blank" rel="noopener">movin_sdk_python</a>
+          <a href="https://github.com/MOVIN3D/MOVIN-SDK-Python" target="_blank" rel="noopener">MOVIN-SDK-Python</a>
           library and a custom
-          <a href="https://github.com/ExperientialTechnologies/Motion-Player-ROS" target="_blank" rel="noopener">ROS 2 retargeting node</a>,
+          <a href="https://github.com/MOVIN3D/Motion-Player-ROS" target="_blank" rel="noopener">ROS 2 retargeting node</a>,
           then trained as reinforcement learning policies using
           <a href="https://github.com/mujocolab/mjlab" target="_blank" rel="noopener">mjlab</a>
           (PPO on GPU-accelerated
@@ -145,13 +145,19 @@
   <section class="deploy reveal">
     <div class="about-inner">
       <h2>PIPELINE</h2>
-      <p class="deploy-intro">The complete motion-to-policy pipeline uses three open-source tools. Every trained policy is available as an ONNX model (160-dim input, 29-dim output) with baked-in observation normalization.</p>
+      <p class="deploy-intro">The complete motion-to-policy pipeline uses five open-source tools. Every trained policy is available as an ONNX model (160-dim input, 29-dim output) with baked-in observation normalization.</p>
       <div class="deploy-options">
         <div class="deploy-option">
           <div class="deploy-num">01</div>
+          <h3>MOVIN Studio</h3>
+          <p>Recording and processing software for MOVIN TRACIN markerless mocap. Captures full-body motion in real-time using LiDAR + vision at 60 FPS. Exports BVH with 51-joint humanoid skeleton, plus FBX for Blender, Maya, Unreal, and Unity.</p>
+          <p><a href="https://www.movin3d.com/studio" target="_blank" rel="noopener">movin3d.com/studio</a></p>
+        </div>
+        <div class="deploy-option">
+          <div class="deploy-num">02</div>
           <h3>Motion-Player-ROS</h3>
           <p>ROS 2 package for retargeting and previewing motion capture on the G1. Supports both playback of pre-recorded <code>.pkl</code> files and real-time retargeting from live MOVIN TRACIN data via OSC/UDP. Dual visualization shows the original human BVH skeleton alongside the retargeted robot motion in RViz.</p>
-          <p><a href="https://github.com/ExperientialTechnologies/Motion-Player-ROS" target="_blank" rel="noopener">github.com/ExperientialTechnologies/Motion-Player-ROS</a></p>
+          <p><a href="https://github.com/MOVIN3D/Motion-Player-ROS" target="_blank" rel="noopener">github.com/MOVIN3D/Motion-Player-ROS</a></p>
           <pre><code># Playback mode
 ros2 launch motion_player player.launch.py \
   motion_file:=clip.pkl bvh_file:=clip.bvh
@@ -161,7 +167,19 @@ ros2 launch motion_player realtime.launch.py \
   port:=9000 human_height:=1.75</code></pre>
         </div>
         <div class="deploy-option">
-          <div class="deploy-num">02</div>
+          <div class="deploy-num">03</div>
+          <h3>video2robot</h3>
+          <p>End-to-end pipeline converting any video to robot motion. Extracts 3D human pose via <a href="https://github.com/AIM-Intelligence/video2robot" target="_blank" rel="noopener">PromptHMR</a> (SMPL-X), then retargets to the G1's 29-DOF joint space using <a href="https://github.com/YanjieZe/GMR" target="_blank" rel="noopener">GMR</a> inverse kinematics. Works with YouTube, phone video, or AI-generated footage — no mocap hardware needed.</p>
+          <p><a href="https://github.com/AIM-Intelligence/video2robot" target="_blank" rel="noopener">github.com/AIM-Intelligence/video2robot</a></p>
+          <pre><code># Extract pose from video
+python scripts/extract_pose.py --project data/clip
+
+# Retarget to G1
+python scripts/convert_to_robot.py \
+  --project data/clip --robot unitree_g1</code></pre>
+        </div>
+        <div class="deploy-option">
+          <div class="deploy-num">04</div>
           <h3>mjlab</h3>
           <p>GPU-accelerated RL training framework combining Isaac Lab's manager-based API with <a href="https://github.com/google-deepmind/mujoco_warp" target="_blank" rel="noopener">MuJoCo-Warp</a> simulation. Trains PPO policies across 8,192 parallel environments on a single GPU. Motion imitation uses 14-body tracking with reward shaping for position, orientation, velocity, and collision avoidance.</p>
           <p><a href="https://github.com/mujocolab/mjlab" target="_blank" rel="noopener">github.com/mujocolab/mjlab</a></p>
@@ -172,7 +190,7 @@ uv run train Mjlab-Tracking-Flat-Unitree-G1 \
   --agent.max-iterations 30000</code></pre>
         </div>
         <div class="deploy-option">
-          <div class="deploy-num">03</div>
+          <div class="deploy-num">05</div>
           <h3>RoboJuDo</h3>
           <p>Plug-and-play sim2real deployment framework for humanoid robots. Modular architecture separates controller (joystick/keyboard/motion), environment (MuJoCo sim or real robot via Unitree SDK), and policy (ONNX/JIT). Supports seamless switching between sim2sim validation and real hardware with minimal code changes.</p>
           <p><a href="https://github.com/GDDG08/RoboJuDo" target="_blank" rel="noopener">github.com/GDDG08/RoboJuDo</a></p>
@@ -285,7 +303,7 @@ python scripts/run_pipeline.py --config=g1_real</code></pre>
         </div>
       </div>
       <div class="equip-block">
-        <h3>Inference</h3>
+        <h3>Training</h3>
         <div class="equip-name"><a href="https://www.dell.com/en-us/shop/desktop-computers/dell-pro-max-desktop-with-nvidia-gb10/spd/dell-pro-max-gb10-desktop/usepmgb10bts" target="_blank" rel="noopener">Dell Pro Max with GB10</a></div>
         <div class="equipment-grid">
           <div class="equipment-item">